Risk and Benefit Analysis of Gain of Function Research - Draft Final Report PDF

Preface: A Brief Guide to Gryphon’s Risk and Benefit Analysis of Gain of
Function Research
The U.S. Government (USG), as part of its continued focus on biosafety and biosecurity, has undertaken
a deliberative process to carefully examine the risks and benefits associated with GOF studies, notably,
those that have the potential to generate pathogens, such as influenza viruses or coronaviruses, with
“enhanced pathogenicity or transmissibility in mammals.” This process, which began in October 2014, is
intended to inform the policy making process regarding future funding decisions by the United States in
this area. The National Science Advisory Board for Biosecurity (NSABB) was tasked with making
recommendations to the USG on this matter and received a two-part charge: first, to make
recommendations on the design, development and conduct of risk and benefit assessments (to be
conducted by an independent, private firm), and second, to make recommendations on a conceptual
approach to the evaluation of proposed GOF studies.
The independent assessment of risks and benefits of GOF research (RBA) was conducted by Gryphon
Scientific. This assessment is one of several sources of information that are being used by the NSABB to
develop its final recommendations. Other key inputs are examinations of the existing policy landscape
and of ethical and decision-making frameworks.
This preface is intended to provide readers with a brief introduction to what is encompassed within the
RBA and to suggest an approach to reviewing it. The RBA starts with a Table of Contents and an
Executive Summary (Chapter 1, page 1), and then consists primarily of two major sections: chapters and
appendices. The Executive Summary is a very high level overview of the results of the analysis of the
risks and benefits of GOF research as compared to research with wild type pathogens and research
based on alternative approaches to GOF studies. It is the entry point for reading and understanding the
RBA. Everyone with an interest in the outcome of the deliberative process is encouraged to read this
part of the RBA.
For those who want to delve further, following the Executive Summary, the following chapters would be
of interest:
• Chapter 2. A detailed introduction and description of the purpose of the RBA

• Chapter 3. Overview of the methods used throughout
• Chapter 4. Background information on influenza, coronaviruses, GOF research
• Chapter 5. Historical information about disease burden and pandemic caused by naturally-
occurring influenza, MERS, SARS
• Chapter 6. Biosafety risk assessment (risks of laboratory acquired infections, local outbreaks,
and global outbreaks for various phenotypes that may be generated by GOF studies)
• Chapter 7. Biosecurity risk assessment (risks of misuse of strains, materials, and other products
of GOF research by various actors; analysis of the systems to prevent misuse)
• Chapter 8. Information risk assessment (risks of misuse of information generated as a result of
GOF studies)
• Chapter 9. Benefits assessment (benefits of GOF research to scientific information,
biosurveillance, countermeasure development, and policy decisions.)
• Chapter 10. Analysis of the potential proliferation of GOF research
• Chapter 11. Analysis of the risk of loss of trust in science
Following these chapters, there is a series of appendices that amplify material in the chapters, or
analyze some related issues.
The majority of the key information about the results of the assessment (biosafety risks, biosecurity risks
and benefits) is presented in Chapters 6, 7, 8, 9. Each of these chapters has a summary at the beginning
(section 6.1, page 78; section 7.1, page 169; section 8.1, page 212; and section 9.1; page 243) which is
then followed by more detailed descriptions of methods, analyses, and results. The chapter summaries
and associated tables and figures are particularly worth examining, and should help readers to
understand the basis for the conclusions in the Executive Summary. The remainder of chapters 6-9
includes much more detailed descriptions of methods, analyses and results.
Those interested in even more thorough descriptions of the background, methods and results of the
RBA analyses, are referred to the other chapters and the appendices. In addition, it is worth pointing
out the inclusion of sections dealing with interesting related matters such as potential proliferation of
GOF research (Chapter 10, page 245) and the globalization potential of GOF benefits (section 9.14, page
423. Finally, there will be available, on line, Supplemental Materials that contain the data that support
the results of the analyses and include methods at the most detailed level for those who may be
interested in reproducing the work.
This preface was produced by the National Institutes of Health, Office of Science Policy.
Risk and Benefit Analysis of
Gain of Function Research
Draft Final Report—December 2015

Risk and Benefit Analysis of Gain of Function Research
This work was conducted under NIH Contract# HHSN263201500002C with Gryphon Scientific from
March 20, 2015 to December 15, 2015.
Susan Apter was the NIH contracting officer and Kelly Fennington was her representative
Project Team:
Gryphon Scientific, Prime Contractor, Rocco Casagrande, Principal Investigator
Signature Science, Subcontractor
Abt Associates, Subcontractor
• Chapter 1, Executive Summary: Corey Meyer, Kavita Berger and Rocco Casagrande
• Chapter 2, Introduction: Rocco Casagrande
• Chapter 3, Overview of Methodology: Corey Meyer, Kavita Berger, Robert Stephan and Rocco
Casagrande
• Chapter 4, Background: Erin Lauer, Mikaela Finnegan, Corey Meyer and Rocco Casagrande
• Chapter 5, Historical Context: Erin Lauer and Rocco Casagrande
• Chapter 6, Biosafety Risk Assessment: Ryan Ritterson, Mark Kazmierczack, Molly Isbell
(Signature Science), Christopher Hulme-Lowe (Signature Science), Erin Lauer, Mikaela
Finnegan, Haley Krem, Audrey Cerles, Gautam Venugopalan, Ryan Handoko, Jacqueline Chu,
Danielle Fields and Rocco Casagrande
• Chapter 7, Biosecurity Risk Assessment of Acts Targeting a Laboratory: Kavita Berger,
Robert Stephan, Phillipe Mauger, Gautam Venugopalan and Rocco Casagrande
• Chapter 8, Biosecurity Risk of GoF Information: Landy Sun, Mikaela Finnegan, Phillipe
Mauger, Craig Hooper and Rocco Casagrande
• Chapter 9, Benefit Assessment: Emily Billings, R. Alex Coots, Ryan Handoko, Phillipe
Mauger, Ryan Ritterson, Rocco Casagrande, Casey Basham and Corey Meyer
• Chapter 10, Proliferation of GoF Research: Luba Katz (Abt Associates) and Rocco
Casagrande
• Chapter 11, Risk of Loss of Trust in Science: Julia Homstad and Rocco Casagrande
Acknowledgements:
The project team would like to acknowledge the guidance and assistance given by the NIH Office of
Science Policy, specifically: Kelly Fennington, Christopher Viggiani, Stuart Nightingale (contractor),
Kevin Ramkissoon, Marina O’Reilly, and Rona Hirschberg (contractor). We would like to thank Allison
Mistry (Gryphon Scientific) and Luba Katz (Abt Associates) for their critical read and editing of the
document. We thank Justin Anderson for his patience with formatting the many drafts of this document.
Also, we thank Barbara Johnson for her guidance on biosafety procedures in containment laboratories and
Gerald Epstein and Wendy Hall of the Department of Homeland Security for a critical review of Chapter
8.
All cover images are from the CDC’s Public Health Image Library: Top left, Microbiologist working at
BSL3-E with H5N1 (image #8675), right, electron micrograph of MERS-CoV (image #18115) and
bottom left, illustration of generic influenza virion (image #11877).
TABLE OF CONTENTS
1 Executive Summary 1
1.1 Biosafety Risk Assessment 1
1.2 Biosecurity Risk Assessment of Malicious Acts Targeting a GoF Laboratory 4
1.3 Biosecurity Risk Assessment of GoF Information 5
1.4 Benefit Assessment of GoF Research 6
2 Overview and Purpose 8

2.1 Organization of the Project 8
2.2 Time Horizon 9
2.3 Interpreting the Results of the RBA 9
3 Overall Methodology 11
3.1 RA of accidents and natural disasters 11
3.2 Biosecurity RA of malicious acts targeting a laboratory 13
3.3 Biosecurity RA of GoF Information 14
3.4 Benefit Assessment 14
4 Background Information on Influenza Viruses, the Coronaviruses and GoF Research 16

4.1 Influenza Viruses 17
4.2 The SARS- and MERS-coronaviruses 26
4.3 An overview of GoF Research 34
5 Historical Context of Outbreaks of Influenza, SARS and MERS 42

5.1 Purpose and Context 43
5.2 Severe Acute Respiratory Syndrome 43
5.3 Middle East Respiratory Syndrome 51
5.4 Influenza 54
6 Risk Assessment of Laboratory Accidents and Natural Disasters 76

6.1 Overview of Results 78
6.2 Methodology 85
6.3 Practices in GoF Laboratories that Reduce Risk but Are Not Included in Our Study 100
6.4 Probability of Laboratory Acquired Infections 105
6.5 Consequences of an Outbreak Caused by an Avian Influenza Strain that Is Not Transmissible in
Mammals 124
6.6 Risk of an Outbreak Escaping Local Control of Pathogens that Are Transmissible in Mammals 126
6.7 Consequences of a Global Pandemic of Pathogens that Are Transmissible in Mammals 139
6.8 Supporting an Estimate of Absolute Risk 161
6.9 Using the Parametric Risk Assessment: Example Calculation 165
7 Biosecurity Risk of Malicious Acts Targeting a GoF Laboratory 168

7.1 Biosecurity Risk Assessment: Summary 169
Findings: Assessment of the Offense (Possible Threats to U.S. Research Laboratories) 171
7.2 Findings: Defense Assessment 178
Risk & Benefit Analysis of Gain of Function Research Gryphon Scientific, LLC i
Analysis of Offense and Defensive Measures 181
8 Biosecurity Risk of GoF Information 211

8.1 Summary 212
8.2 Purpose and Approach 214
8.3 Methods 214
8.4 Baselining the Biological Threat 216
8.5 Overview of the State of the Science of Dual Use GoF Information 228
8.6 Evaluation of the Capability and Intent of Malicious Actors to Leverage Dual Use Information 237
9 Benefit Assessment of GoF Research 240

9.2 Methodology 255
9.3 Coronaviruses: Benefits of GoF research 263
9.4 Introduction to GoF Research Involving Influenza Viruses 286
9.5 Influenza viruses: Benefits of GoF Research that Enhances Virus Production 289
9.6 Influenza viruses: Benefits of GoF research that enhances mammalian adaptation and
transmissibility 301
9.7 Influenza viruses: Benefits of GoF research that enhances virulence 329
9.8 Influenza viruses – Benefits of GoF research that leads to Evasion of Existing Natural or
Induced Adaptive Immunity 351
9.9 Influenza viruses: Benefits of GoF Research that Leads to Evasion of Vaccines 371
9.10 Influenza Viruses: Benefits of GoF Research that leads to Evasion of Therapeutics 373
9.11 Influenza viruses: Benefits of GoF Research Involving Reassortment 393
9.12 Evaluation of the quantitative benefits of GoF research 403
9.13 Likelihood of GoF Strains Arising in Nature 408
9.14 Evaluation of the Globalization Potential of GoF Research 423
10 Potential Proliferation of GoF Research 445

10.1 Summary 446
10.3 Methods 446
10.4 Limitations 448
10.5 Findings 449
10.6 Conclusions 462
11 Risk of Loss of Trust in Science 464

11.1 Summary 464
11.3 Results 466
12 Appendix I: Glossary 480
13 Appendix II. Acronyms Used 486
14 Appendix III. Additional Data on the Methods of the Quantitative Risk Assessment 489
14.1 Additional Methodological Information Supporting the Estimate of Loss of Containment
Pathways 490
Risk & Benefit Analysis of Gain of Function Research Gryphon Scientific, LLC ii
14.2 Methodological Details of the Branching Process Model 506
14.3 Methodological Details of the HHS-BARDA IIM 512
14.4 Additional Data on the Potential Proliferation of GoF Research 514
15 Appendix IV. Benefit Assessment 518

15.1 Coronaviruses: Detailed Analysis of the Benefits of GoF Research 520
15.2 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances Virus
Production 565
15.3 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Enhances
Mammalian Adaptation and Transmissibility 596
15.4 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances Virulence 651
15.5 Influenza viruses: Detailed analysis of the benefits of GoF research that leads to evasion of
existing natural or induced adaptive immunity 693
15.6 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Leads to Evasion of
Vaccines 728
therapeutics 730
15.8 Influenza viruses: Detailed analysis of the benefits of GoF research involving reassortment 760
15.10 List of Subject Matter Experts Interviewed for the Benefit Assessment 821
16 Appendix V: Findings Informing the Biosecurity Risk Assessment 826

16.1 Purpose of the Biosecurity Risk Assessment (RA) 828
16.2 Methodology 829
16.3 Definitions of Terms Used in the Threat Matrix 839
16.4 Analysis of Malicious Actor Capabilities and Motivations 845
16.5 Analysis of Historical Incidents 851
16.6 Attacks against Laboratories 877
16.7 Biocrimes Committed by Individuals 881
16.8 Terrorist and Extremist Events Tied to Biological Warfare 888
16.9 Designated Foreign Terrorist Organizations and Biological Weapons 896
16.10 Detailed History of Known Terrorist Biological Weapons Programs 903
16.11 Other Terrorist/Extremist Groups Linked in Some Fashion to Biological Weapons 915
16.12 The Islamic State of Iraq and the Levant (ISIL) 917
16.13 Biosafety and Biosecurity at U.S. Research Laboratories 920
16.14 Laws, Guidance, Policies, Practices, and International Agreements on Biosafety and Biosecurity928
16.15 Restriction of Fundamental Research, Dual Use Research of Concern and Recombinant DNA
Guidelines 969
16.16 Analysis of Security Measures: Requirements, Implementation, and Gaps of Security Measures 973
Risk & Benefit Analysis of Gain of Function Research Gryphon Scientific, LLC iii
1 Executive Summary
The analysis described in this report, provides information for the NSABB to make recommendations
about a general conceptual approach to the evaluation of Gain of Function studies, and for the US
government to formulate policy regarding Gain of Function (GoF) research. In this document, the term
GoF is used in the same manner as the Framework for Conducting Risk and Benefit Assessments of Gain
of Function Research—the “Framework”. By design, this study was broad in its scope, intentionally
assessing all of the traits and pathogens mentioned in the Framework to determine where risk lies. The
conclusions of the risk assessment should point to the pathogens and the enhanced phenotypes that would
increase risk of a pandemic and those that do not increase this risk. Similarly, the benefit assessment
determines which experiments (regardless of their risk) have important and unique benefits.
This project is divided along the three major tasks, each of which requires a distinct data collection and
analysis approach: 1) a risk analysis (RA) of accidents and natural disasters, 2) a biosecurity RA and 3) a
benefit assessment. The RA of accidents and natural disasters (called the biosafety RA for simplicity)
leveraged sophisticated quantitative modeling of the probability and consequences of various events that
lead to an outbreak, the ongoing transmission of the outbreak in humans and the termination of the
outbreak by public health measures or natural forces. The biosecurity RA requires an analysis of data
from the intelligence and law enforcement community on malicious actors and an assessment of the
efficacy of security measures at preventing or mitigating a hostile act. The biosecurity RA is delivered in
two parts because risks posed by malicious acts targeting laboratories that conduct GoF required a
different analytical approach than the assessment of the risk generated by the misuse of published GoF
research. The benefit assessment requires an understanding of the gaps in scientific knowledge, public
health and medicine that GoF experiments could address. Moreover, this assessment requires the
identification of scientific and non-scientific barriers to the realization of these benefits.
1.1 Biosafety Risk Assessment
The Biosafety Risk Assessment evaluated the increase in risk to human health of pandemics caused by
modified strains of the influenza viruses and the coronaviruses. This RA uses the word “coronavirus” to
mean the coronaviruses that cause SARS or MERS and not the coronaviruses that cause the common
cold. In every case, the increase in risk compared to wild type strains was provided to determine if GoF
experiments could create pathogens that are more likely to cause laboratory acquired infections, to create
a local outbreak or to cause a global outbreak of greater consequence than those strains that evolved via
natural forces. Note that although this study identified several types of risky, theoretical GoF experiments,
many of these experiments have not been described in the literature. For example, no examples of
researchers endeavoring to determine if seasonal influenza viruses could be made more transmissible
were found. Moreover, some GoF studies are performed in highly attenuated strains, so that even though
the risk of an outbreak increases if these strains were modified, risk is increasing from a very low level
toward the level posed by wild type strains.
In short, a strain of influenza virus that is as transmissible (or to which the population has as little
minimal immunity) as newly emerged pandemic strains WHILE producing a disease to a case fatality rate
of more than 5%, would pose more of a risk of a global pandemic than any wild type strain heretofore
identified. No experiments that are likely to be conducted under the rubric of GoF research will drive risk
more than this combination of traits or significantly increase the risk of a laboratory acquired infection.
All other combinations of traits would lead to pathogens that have a lesser total risk than the wild type
1918 strain. Increasing the transmissibility of the coronaviruses, while significantly increasing the risk of
GoF RBA Report Gryphon Scientific, LLC 1

work with those pathogens, still creates a pathogen that poses less of a risk of a global pandemic than the
wild type 1918 influenza strain.
Another major finding of this risk assessment is that only a small minority of loss of containment events,
which are rare in themselves, lead to a global pandemic. Only 0.5% of laboratory associated infections of
seasonal influenza would seed a global pandemic, even assuming the accident was with a strain that has
not circulated recently. Moreover, in the case of seasonal influenza, the risk of a global pandemic is
exacerbated by laboratory research only if that laboratory is working on a strain that has not circulated
recently. If the strain is currently in circulation, the spread of the outbreak is likely to be driven by
travelers, not by laboratory accidents. If the released strain circulated recently, residual immunity is likely
to curtail its spread. Similarly, only 1.5% of laboratory associated infections with wild type pandemic
influenza strains would seed a global pandemic. Wild type strains of avian influenza and the diseases
caused by the coronaviruses are insufficiently transmissible to have a significant chance of causing a
global pandemic.
Because seasonal influenza viruses are associated with a low case fatality rate, GoF experiments that
increase this rate could significantly increase the global death toll from an outbreak, increasing risk.
Developing seasonal influenza strains that are more transmissible than wild type strains (approximately as
transmissible as pandemic strains) or that overcome residual immunity increase the probability that an
outbreak would escape local control and increase the consequences should a global outbreak be initiated
(in terms of both morbidity and mortality). The creation of an antiviral resistant strain could increase the
consequences of a global outbreak, but only in more economically developed countries where caches of
these antivirals could be handed out to a significant fraction of the infected population. A strain of
seasonal influenza that can overcome protective vaccination could also increase the consequences of an
outbreak in high income countries, which has the resources to vaccinate their population quickly.
However, this phenotype is of concern only if immune evasion is afforded by means other than changing
its antigenic properties, which is not a subject of current research in influenza. An unresolved question
(which likely depends on the biology of the virus released and its similarity to currently circulating
strains) is if the laboratory-associated outbreak of seasonal influenza would replace the annual toll of
seasonal influenza by supplanting circulating strains or would it add to this toll.
If GoF strains of seasonal influenza were manipulated at the BSL3 instead of the BSL2 level, risk overall
may not increase much compared to work on wild type strains at BSL2. That is, the rate of laboratory
acquired infections is likely to decrease by three-fold, whereas any GoF phenotype (except for large
increases in pathogenicity) only modestly more than this amount.
In contrast to the several GoF manipulations that could increase the risk posed by seasonal influenza
strains, the only GoF trait that would create a strain of pandemic influenza that poses more risk of a global
outbreak than the most pathogenic, wild type strain (in this case, the 1918 pandemic strain) is the trait of
antiviral resistance, and the increase is modest. That being said, even this trait increases the consequences
in high income countries only because poor countries lack access to a sufficient quantity of antivirals to
effectively protect their populations. GoF manipulations in other strains of pandemic influenza could
increase their risk of causing a global outbreak, but would still match the risk posed by the wild type 1918
pandemic strain. The enhanced growth phenotype COULD increase the risk of a laboratory acquired
infection, if samples with 1E9pfu/ml or 1E10pfu/ml were routinely manipulated in a laboratory
(increasing risk by up to six-fold, respectively). However, it is uncertain if this phenotype is desirable or
even achievable.
Wild type avian influenza is insufficiently transmissible amongst people so that it cannot cause a global
outbreak driven by the spread of the disease among humans. For this reason, no loss of containment event
would lead to a global outbreak from a wild type strain. Because wild type strains of avian influenza

cannot spread globally between people, the creation of strains that are human transmissible would greatly
increase the risk that such an outbreak could occur, which could cause millions of illnesses. The creation
of a strain that is as transmissible as seasonal influenza would have a significant chance of sparking a
global outbreak if a local outbreak were initiated. Assuming that the case fatality rates of the most
pathogenic strains of avian influenza are inflated by the underreporting of mild illness in people,
increasing the pathogenicity in humans could increase the consequences modestly. Adapting avian strains
to humans without increasing transmissibility (thereby lowering the median infectious dose in people)
actually decreases risk because while this manipulation increases in the probability that a single
laboratory worker would become infected, it decreases the risk that birds would become infected through
an accidental release via the solid waste stream, which could lead to thousands of human infection. No
other GoF increases the risk posed by avian influenza.
Similarly, most estimates of the transmissibility of the coronaviruses consider these pathogens to be
insufficiently transmissible and sufficiently susceptible to control measures such that a global pandemic
has a very minimal chance of occurring. For this reason, increasing the transmissibility of the
coronaviruses could significantly increase the chance of a global pandemic due to a laboratory accident.
Because SARS-CoV is more transmissible than MERS-CoV, a relatively modest increase in
transmissibility of SARS-CoV could increase risk, whereas MERS-CoV must be made significantly more
transmissible to drive risk. That being said, even if these strains were modified to be as transmissible as
pandemic influenza, the susceptibility to control measures of the outbreaks they cause would still contain
a majority of the outbreaks initiated. Increasing the pathogenicity of these strains could also increase risk
somewhat through the increase in global deaths expected, especially since most deaths from wild type
strains are suffered by those with significant co-morbidities. However, if a coronavirus were modified
such that it caused a global pandemic, their long incubation time and disease course lead to a pandemic
that unfolds over many years. The fact that the outbreak evolves slowly gives public health authorities
more time to adapt and expand their efforts to further contain the outbreak than the modeling conducted
in this assessment suggests. If a strain with enhanced growth properties were developed and samples with
1E9pfu/ml or 1E10pfu/ml were routinely manipulated in a laboratory, the risk of a laboratory acquired
infection in a coronavirus laboratory would increase by up to ten-fold, respectively. However, it is
uncertain if this phenotype is desirable or even achievable given that the wild type coronaviruses grow
sufficiently well in culture.
The laboratory features and practices that most influence risk include the strict adherence to incident
reporting and isolation protocols for laboratory workers. Minimizing the chance that a worker would
violate either of these protocols can decrease the risk that an infected laboratory worker would create an
outbreak by up to seven-fold when working with seasonal influenza virus or by ten-fold with the
coronaviruses. Additionally, when working with the coronaviruses (which are more stable in the
environment than the influenza viruses), protocols to minimize the hazard posed by the contamination of
the hands (proper use of double-gloving and thorough hand-washing) can also reduce the probability of
an infection by nearly fifty-fold. The probability that workers themselves commit errors that generate the
laboratory accident obviously also drive the probability of an infection by one-hundred-fold or more.
While this conclusion is self-evident, it underscores how extensive worker training prior to entry into a
BSL3 laboratory, the assignment of highly trained workers for critical safety tasks (such as the operation
of autoclaves) and the identification and re-training of careless workers could all significantly improve
safety.
The state of knowledge of the rates and consequences of human errors in life science laboratories is too
poor to develop robust predictions of the absolute frequency with which laboratory accidents will lead to
laboratory acquired infections. Using historical incidents (and lack thereof) as a guide, a rate (at the 90%
confidence level) of a laboratory acquired infection every three to 8.5 years can be set across the 100 or so
laboratories that study influenza and the coronaviruses in the US. Given that this study predicts that 0.4%

of these infections would lead to a global pandemic (since most of these laboratories study seasonal
influenza, and not pandemic influenza), work with wild type influenza viruses would lead a global
pandemic once every 750-50,000 years. A significant risk of an outbreak would be caused only if the
strain released in the accident were a seasonal influenza strain that has not recently circulated, however,
this outbreak could lead to up to 4,000,000 deaths worldwide, but it is uncertain if these deaths would
supplement or supplant the yearly death toll from seasonal influenza. Conservatively, an infection with a
pandemic influenza strain could be expected to lead to a global pandemic once every 560-13,000 years,
causing up to 80,000,000 deaths if the strain used were as pathogenic as the 1918 pandemic strain. Given
that viruses were characterized much less than 100 years ago, it cannot be stated with certainty that these
pathogens will be studied under similar containment conditions far enough into the future for an accident
to be likely to occur even once. Avian influenza strains and coronavirus strains are insufficiently
transmissible to cause a global pandemic.
If well-funded, GoF research could be conducted by up to approximately 40 research groups in the US

because these groups have been performing, or have the capacity to perform, certain types of GOF
experiments involving influenza, MERS, and SARS viruses. This maximum number is supported by the
case studies examined which showed that a new discovery in virology may proliferate to as few as one
and as many as 70 new groups around the world within 10-15 years.
1.2 Biosecurity Risk Assessment of Malicious Acts Targeting a GoF Laboratory
The purpose of this component of the biosecurity risk assessment is to provide NSABB with an
assessment of the increased human health risk posed by a malicious act involving a GoF strain of the
influenza- or coronaviruses compared to malicious acts involving wild type strains. The risk assessment
involved five steps: 1) characterization of the threat, which includes an evaluation of historical incidents
and malicious actor motivation and capability (the “offense”); 2) review of the current security policies
and practices landscape that governs research with influenza, SARS-CoV, and MERS-CoV in the United
States (the “defense”); 3) identification of plausible threats based on analysis of the “offense” and
“defense”; 4) assessment of the potential for the plausible threats to cause infections in the local
community or broader; and 5) comparison of possible pandemic consequences of plausible threats
involving GoF viruses and non-GoF viruses.All of the data collected were used to assess the plausible
threats facing laboratories that perform GoF research. These plausible threats serve as the most probable
events that could lead to a loss of containment from a biosecurity incident. Therefore, they were used to
focus the quantitative analysis of local and widespread infections on those acts that are the most plausible
in today’s laboratory security environment.
Based on historical incidents and an assessment of the security governance in the United States, the most
likely malicious acts to be carried out in or on a containment laboratory include theft of virus stocks,
experimental samples, equipment, or research animals; deliberate contamination of personal protective
equipment or laboratory equipment of co-workers; deliberate compromise of the personal protective
equipment or laboratory equipment of co-workers; and mixing of infected with uninfected samples or
animals outside proper containment. In addition, incidents involving bombs or active shooters may cause
loss of containment if carried out inside or near the entrance of high containment laboratories in which
GoF research is conducted.
In today’s regulatory and security environment, the most plausible malicious acts taking place at high
containment, research laboratories, involves malicious insiders who have authorized access to the
laboratories and virus(s) contained therein. Insiders may work alone or in coordination with an outside
group. Their motivations range from emotional disturbances to ideological radicalization by domestic and
transnational terrorist organizations. The likelihood that outsiders could gain access to a laboratory

without insider assistance is low. Therefore, outsiders present a threat to the periphery of the research
complex or building only, but not a significant threat to the high containment laboratory itself.
Only a handful of GoF traits significantly increase biosecurity risk after a malicious event targets a
laboratory. For seasonal and pandemic influenza, the ability to overcome protective vaccination and
antiviral resistance modestly increases risk by increasing the potential consequences in high income
countries. There is no significant effect on risk if the global population is considered as a whole. No other
trait influences the risk significantly of pandemic influenza strains. For seasonal influenza, increasing the
transmissibility and ability to evade residual immunity significantly increases risk because outbreaks are
more likely to occur, to escape local control and will create more consequential global outbreaks. For
avian influenza, increasing transmissibility greatly increases risk because this modification is required to
spark a global outbreak of a disease by human-to-human contact, potentially infecting millions. Without
this change, the hazard is restricted to those exposed to contaminated materials and infected birds,
limiting the outbreak to thousands of cases at most. Increasing pathogenicity can modestly increase risk.
Similarly, the wild type coronaviruses have a very small chance of sparking a global outbreak so
increasing transmissibility greatly increases risk. Increasing pathogenicity can modestly increase risk.
When comparing the biosafety and biosecurity risks, a successful event that covertly infects the public
(theft from an influenza laboratory of an infected animal, contaminated piece of equipment or viral stock)
must occur once every 80-5,500 years for biosecurity event to have the same total risk as biosafety events.
Given the frequency with which these malicious acts have occurred in the past, this analysis suggests that
biosecurity considerations be given as much weight as biosafety issues.
1.3 Biosecurity Risk Assessment of GoF Information
The biosecurity RA of GoF information is based on the open-source literature covering desirable
characteristics of biological agents and the scientific literature on GoF studies and non-GoF studies with
significant dual-utility. The potential biosecurity information risk that could be generated by GoF
information was assessed compared to what could be achieved through dual-use studies that do not rely
on GoF research. It was then determined if the unique dual-use information resulting from GoF studies
had already been published.
Little information risk remains from GoF research. Although the development of a highly-contagious,
highly virulent strain of influenza presents significant biosecurity information risk, the methods to
produce these strains have already been published and so no information risk remains. Moreover, the
specific changes in the genome that lead to these traits have also been characterized and published, so an
actor could reproduce the dual-use strains using reverse genetics. Similarly, information on how to
develop strains of influenza viruses that grow well in culture/eggs or evade medical countermeasures or
diagnostics has some dual-utility, but the methods to create these strains also have already been
published. A modest information risk would be realized if researchers published methods to produce
strains of influenza viruses that can produce more prolonged or chronic illness. Although this
manipulation is a possible enhancement of pathogenicity that can fall under the definition of GoF
research, there is little scientific rationale to undertake these experiments. Hence, the possibility that this
information risk will be realized is low. Another modest information risk inheres in the publication of
methods to produce strains of influenza virus that are able to overcome protective vaccination even if the
vaccine matches the serotype of the pathogen. Similar work has been published for other pathogens, but
these pathogens have larger and more plastic genomes than the influenza viruses so it is not known if
similar manipulations could be successfully carried out in the influenza viruses.

Significant information risk would be realized by the publication of methods to create a highly
transmissible SARS- or MERS-coronavirus that maintains its pathogenicity. Notably, without an animal
model of transmissibility for these pathogens, this information risk is unlikely to be realized in the near
future. A modest information risk inheres in methods to manipulate the genomic targets of a diagnostic
assay for coronavirus infections without compromising the other desirable traits of the pathogen.
State actors (and the sub-state groups they sponsor) are currently the only groups with the resources,
expertise, motivation and time to leverage this dual-use information. These states could protect their own
populace from a global pandemic by secretly stockpiling vaccines that are protective against their
modified strain. For this reason, states would be more likely to produce modified influenza viruses than
coronaviruses (because no vaccines exist for this type of agent) and would probably be uninterested in
developing strains able to overcome any vaccine (as this strain would vitiate their comparative
advantage). Sub-national malicious actors may obtain the capability to replicate some of the less complex
GoF studies, but have so far not demonstrated any capacity to work with viral agents and little capacity
for waging biological warfare in general. Highly skilled individuals trained in biology would be capable
of replicating GoF studies, but are currently constrained greatly by a lack of material resources and time.
Finally, no information risks unique to GoF research were identified. Similar techniques to those used in
GoF experiments could be leveraged for other pathogens that are not captured by the moratorium and are
therefore outside the GoF framework to create a highly transmissible strain of an already deadly virus
(like the Hendra and Nipah viruses) or to create a deadly strain of an already highly transmissible
pathogen that has been modified to overcome protective vaccination (polio-, mumps- or measles-virus).
Perhaps most worryingly, reverse genetics techniques could be used to synthesize smallpox virus if an
actor has significant molecular biology skill, and this strain could be modified to overcome protective
vaccination. Non-GoF pathogens could be used to produce effective, novel incapacitating agents by the
modification of a highly contagious virus (polio-, mumps- or measles-virus) to overcome protective
vaccination.
1.4 Benefit Assessment of GoF Research
The benefit assessment describes the potential benefits of GoF research involving influenza viruses and
coronaviruses, relative to two different types of alternative approaches: alternative experimental
approaches that can provide the same or similar information, and alternative scientific or technical
innovations that may similarly benefit public health through completely different mechanisms. Notably,
this assessment is limited to the evaluation of GoF experiments that have been published in the scientific
literature.
Within the field of CoV research, GoF approaches in the following phenotypic categories were identified:
enhanced pathogen production, altered host range, enhanced virulence, and evasion of therapeutics in
development. GoF approaches that alter host range and enhance virulence uniquely enable the
development of animal model systems that recapitulate human disease pathogenesis, which are critical for
establishing the safety and efficacy of candidate vaccines and therapeutics and for the study of disease
pathogenesis mechanisms. GoF approaches that enhance virulence are also uniquely capable of
demonstrating that live attenuated vaccines (LAVs) do not recover virulence upon growth in vivo, an
important aspect of safety testing of candidate LAVs. Of note, this particular type of experiment simply
increases the human health risk of the attenuated strain to approach that of wild type strains. GoF
approaches that enhance virulence represent the most efficient and effective strategy for discovering
novel virulence factors, which may be good targets for new therapeutics. However, several alternative
strategies for the development of new therapeutics are being actively pursued and have also shown
promise. GoF approaches that lead to evasion of therapeutics are critical for the development and

regulatory approval of new therapeutics. Because these therapeutics are not yet widely available, no
increase in human health risk is posed by resistant strains. GoF approaches that alter host range and
enhance virulence provide unique benefits to study cross-species adaptation and pathogenicity, but
alternative approaches may also be used. Of note, this adaptation to a new host typically attenuates
virulence in the original host (in the case of SARS and MERS-CoV, humans).
Within the field of influenza research, GoF approaches in the following phenotypic categories were
identified: enhanced pathogen production, mammalian adaptation and enhanced transmissibility,
enhanced virulence, evasion of vaccines or therapeutics, and evasion of existing natural or induced
immunity. Across all GoF phenotypes, GoF approaches provide unique benefits to the study of the
mechanistic basis of the phenotype under study as well as the evolutionary mechanisms driving
acquisition of that trait, though alternative approaches may also be used. Alternative approaches have
stringent limitations for the study of mechanisms underlying mammalian transmissibility of animal
influenza viruses, as animal flu viruses that efficiently transmit in humans do not exist in nature. GoF
approaches that enhance virus production are uniquely critical for the current ability to produce sufficient
and timely influenza vaccines for seasonal flu epidemics and flu pandemics and represent the only
strategy for improving existing vaccine production capabilities in the near-term. Of note, GoF approaches
used in vaccine production attenuate an otherwise pathogenic strain while enhancing its growth
properties. GoF approaches that enhance the infectivity, transmissibility, and virulence of animal flu
viruses inform pandemic risk assessments of circulating influenza viruses, which guide downstream
decision-making about investments in pre-pandemic vaccine development and other pandemic
preparedness initiatives. In general, GoF research moderately contributes to the overall risk associated
with a particular virus. However, GoF data play an important role in rapid risk assessments when novel
flu viruses first emerge in human populations due to the early availability of sequence data. These risk
assessments facilitate more rapid initiation of response activities such as pre-pandemic vaccine
development. Of note, realization of this benefit is subject to significant advancements in the state of
knowledge about mechanisms underlying mammalian adaptation, transmissibility, and virulence.
Information derived from these GoF approaches also guide selection of viruses used as the basis of pre-
pandemic vaccines. GoF approaches that enhance the infectivity and virulence of influenza viruses are
also used to develop animal models that support the study of disease pathogenesis and medical
countermeasure (MCM) development. GoF approaches that lead to evasion of therapeutics are critical for
the development and regulatory approval of new therapeutics. Because these therapeutics are not yet
widely available, no increase in human health risk is posed by resistant strains. However, similar
approaches using licensed therapeutics inform therapeutic recommendations for seasonal influenza
infections and pandemic preparedness initiatives for high-risk animal influenza viruses, but phenotypic
approaches for antiviral sensitivity testing are also used for these purposes. GoF approaches that lead to
evasion of vaccines are uniquely capable of determining whether viruses can acquire mutations to escape
neutralization of candidate broad-spectrum or universal influenza vaccines, a critical aspect of testing the
potential field efficacy of vaccines in development. Because these vaccines are not yet widely available,
no increase in human health risk is posed by strains that can overcome the protection afforded by these
vaccines. GoF approaches that lead to evasion of existing natural or induced immunity have potential to
improve the efficacy of seasonal influenza vaccines, but this benefit is subject to advancements in the
state of knowledge about the mechanistic basis of antigenic drift as well as expansion of sequencing
capabilities across public health laboratories involved in global influenza surveillance. Finally, GoF
studies involving reassortment, which may lead to one or more phenotypic changes, are uniquely capable
of providing information that can be used to prioritize community-level interventions aiming to prevent
opportunities for co-infections that could lead to the generation of reassortant viruses with phenotypic
properties of concern.

2 Overview and Purpose
The overarching purpose of conducting the risk/benefit analysis (RBA) is to provide information for the
NSABB to make recommendations about a general conceptual approach to the evaluation of Gain of
Function studies, and for the US government to formulate policy regarding gain-of-function (GoF)
research. In this document, the term GoF is used in the same manner as the Framework for Conducting
Risk and Benefit Assessments of Gain of Function Research—the “Framework”.1 By design, this study
was broad in its scope, intentionally assessing all of the traits and pathogens mentioned in the Framework
to determine where risk lies. The conclusions of the risk assessment should point to the pathogens and the
enhanced phenotypes that would increase risk of a pandemic and those that do not increase this risk.
Similarly, the benefit assessment determines which experiments (regardless of their risk) have important
and unique benefits. That being said, this study was not so broad as to assess the risk posed by
experiments that could create pandemic pathogens that do not all within the Framework. Specifically, as
discussed in Chapter 8, other pathogens that lie outside the framework could be manipulated to cause a
global outbreak. Also, other traits of the influenza viruses and coronaviruses could be manipulated that
would alter their pandemic potential (such as their environmental stability, which could significantly
increase their risk of causing a laboratory acquired infection, and the probability that patients could
transmit the disease to others prior to the onset of symptoms).
The specific goals of this assessment is to provide evidence on how particular GoF experiments affect the
following possibilities:
• That an outbreak caused by a laboratory accident may occur,

• That an outbreak caused by a laboratory accident may increase in severity or extent,
• That a hostile actor may misuse the materials or information generated,
• That future or ongoing disease outbreaks or attacks could be prevented or mitigated, and
• That the life-science research in general would be advanced.
2.1 Organization of the Project
This project is divided along the three major tasks, each of which requires a distinct data collection and
leveraged sophisticated quantitative modeling of the probability and consequences of various events that
outbreak by public health measures or natural forces. The biosecurity RA requires an analysis of data
from the intelligence and law enforcement community on malicious actors and an assessment of the
efficacy of security measures at preventing or mitigating a hostile act. The biosecurity RA is delivered in
two parts because risks posed by malicious acts targeting laboratories that conduct GoF required a
different analytical approach than the assessment of the risk generated by the misuse of published GoF
research. The benefit assessment requires an understanding of the gaps in scientific knowledge, public
health and medicine that GoF experiments could address. Moreover, this assessment requires the
identification of scientific and non-scientific barriers to the realization of these benefits.
1 Framework for Conducting Risk and Benefit Assessment of Gain-of-Function Research: Recommendations of the National
Advisory Board for Biosecurity. May 2015,
http://osp.od.nih.gov/sites/default/files/resources/NSABB_Framework_for_Risk_and_Benefit_Assessments_of_GOF_Resea
rch-APPROVED.pdf

2.2 Time Horizon
The life sciences are advancing extremely rapidly such that techniques in common use today were
unheard of years ago, and the findings are being applied to more and more facets of life. The state of the
science many decades hence is impossible to predict. To ground the work in real science, the RBA is
constrained by a five year time horizon. This time horizon is necessary because the approach is data
driven and the future state of research protocols is unknowable, especially how these changes will affect
stocks of pathogen, containment measures, public health measures and gaps in scientific knowledge,
public health and medicine. New modes of scientific inquiry could obviate GoF research or open up new
opportunities for its application. New laboratory techniques could greatly reduce the chance that an
accident would occur or that any infections may happen. Of relevance to biosecurity, the malicious actors
who may misuse the fruits of GoF research in the far future may have motives or capabilities much
different from those of today’s actors.
Specifically, all risks are considered in a five year time horizon. In contrast, the follow-on benefits of a
scientific discovery that is produced in a five year time-frame will be considered even if these benefits are
realized further into the future. This expanded time-horizon for benefits is necessary because basic
science finds its application in the field years after a discovery in basic science and some regulatory
processes require more than five years by themselves before products borne of a scientific discovery can
be used.
2.3 Interpreting the Results of the RBA
In this study, GoF phenotypes are analyzed individually so that the NSABB can understand how any
particular anticipated change would affect risk in isolation. In reality, many of the phenotypes considered
by the framework are inextricably linked. For example, a component of transmissibility of seasonal
influenza in human populations is the protection afforded by exposure to similar strains in the past. For
this reason, the ability to overcome residual immunity influences transmissibility. Similarly, adaptation to
a host is a necessary component of being transmissible in that host. A strain that is adapted to a host is
likely to grow to a higher titer in cells derived from that host, and produce a higher titer infection in a
living host. High titer infections may often lead to a greater amount of viral shedding and so these
phenotypes are likely related to transmissibility.
The modeling completed enables a complete assessment of how any combination of parameter values that
describe the pathogen and control measures influences risk, however, all possible combinations of these
values and their influence on risk cannot be shown concisely in a report. Instead, static slices through this
very complex risk space are taken and shown as two-dimensional figures in this report that explore the
effect of changing one parameter while allowing all others to vary.
This study examines the risk should a GoF experiment lead to a pathogen with particular traits. In a
quantitative framework, these traits must be described numerically (such as a specific increase in the
reproductive number of the outbreak or the median infectious dose). However, quantitatively translating
empirical studies of transmission in animals to epidemiological predictions for human populations is
impossible. That is, increases in transmissibility in ferrets in isolators cannot be linked to a specific
increase in the reproductive number for outbreaks in human populations. Therefore, it is unknown if the
enhanced transmissibility observed in GoF experiments done to date would significantly change the risk
of an outbreak. Only one component of the transmissibility of a virus in a human population is the
biology of the virus and the host because humans may change their behavior to reduce the risk of contact
during a particularly worrying outbreak. In fact, a recent study estimates that the ferret model of influenza

can be used to explain only 66% of the variation in transmissibility in humans observed across subtypes.2
Instead, this RBA can simply determine how particular increases in transmissibility of a pathogen causing
a human outbreak would influence risk. The feasibility of achieving any particular phenotype via GoF
research is a question of science.
Lastly, this study uses the actuarial definition of risk (risk is the product of probability and consequences
of a bad event). Wherever possible, this study clearly describes how aspects of GoF research influence
risk by increasing the probability that an outbreak would occur and/or by increasing the consequences
should it occur. In this way, readers can use this document to inform their calculations based on other
possible definitions of risk (the probability that a bad event of any consequence occurs, for example).
2 Buhnerkempe, MG et al, “Mapping influenza transmission in the ferret model to transmission in humans” eLife, 2015,
e07969.

3 Overall Methodology
This project is divided along three major tasks, each of which requires a distinct data collection and
requires sophisticated quantitative modeling of the probability and consequences of various events that
outbreak by public health measures or natural forces. The biosecurity RA, which considers acts that
originate in a GoF laboratory and the misuse of the information generated by GoF research, requires an
analysis of historical data on malicious actors and an assessment of the efficacy of security measures at
preventing or mitigating a hostile act. The benefit assessment requires an understanding of the gaps in
scientific knowledge, public health and medicine that GoF experiments could address. Moreover, this
assessment requires the identification of scientific and non-scientific barriers to the realization of these
benefits.
3.1 RA of accidents and natural disasters
The RA of accidents and natural disasters must provide risk information about research that has yet to be
initiated in locations that have yet to be identified if this assessment is to inform a system for the
evaluation of future research. This RA must also consider work with wild type pathogens that does not
fall under the umbrella of GoF research. Unfortunately, the experiments to manipulate pathogens with
pandemic potential (PPP), the resultant phenotypes, the biosafety features of the laboratory and the
environment around the laboratory all could significantly influence risk. To cover the entire landscape of
experiments, phenotypes, containment measures and environments, we took a parametric approach to risk
modeling. That is, we determined how changing any attribute of the pathogen, experiment, laboratory or
environment would affect risk and then bound this assessment in science by assigning real examples to
particular values. For example, we assessed how the transmissibility of an influenza virus affects risk of
an outbreak from arbitrarily small values of transmissibility through arbitrarily large values. In this
manner, we provide information on how transmissible an influenza virus must be in order for risk to
increase significantly. We then compared this “break point” to the transmissibility of known influenza
viruses to provide context on the feasibility of novel strains reaching this level of transmissibility. A
similar approach was taken with all GoF phenotypes. Similarly, biocontainment aspects and features of
the environment were explored for their influence on risk and we highlight those features or qualities of
the environment that may significantly influence risk.
Using this approach, we considered biosafety risk by its component parts: the probability that an event
would occur that would lead to an infection outside the laboratory, the probability that the infection
would lead to an outbreak that seeds a global epidemic, and the consequences of the global epidemic.
The RA of accidents and natural disasters began with the accidents and natural disasters themselves. Of
all the events that COULD lead to a loss of containment that could befall a laboratory, we chose to
quantitatively model those that were either identified as high-risk in previous laboratory risk assessments,
cited as frequent causes of accidents in laboratories in incident reports or those that are uniquely relevant
to GoF studies. These events included high-probability, low-consequence events (like spills), low-
probability, high-consequence events (like earthquakes) and “maximum reasonably foreseeable events”.
Events that are both low-probability and low-consequence were considered but not modeled further
because they will, by definition, not contribute to risk.

Because of the routine use of Fault Tree Analysis (FTA) in a Probabilistic Risk Assessment (PRA)
framework in the estimation of risks arising from accidents and natural disasters creating technological
hazards (accidents and disasters striking nuclear power plants and the chemical supply chain, for
example),3 we applied a similar methodology here. In FTA, the probability that a specific hazard is
generated via a series of connected failures is estimated. This analysis method is most commonly used to
understand how systems can fail and to identify the best ways to reduce risk. To explore the uncertainty in
parameter values and the variety of possible paths through a fault tree, we employed Monte Carlo
simulations, in which repeated, random draws of possible paths and parameter values are sampled to
obtain an aggregate realization of risk. For each incident (and for all incidents in aggregate) we obtained a
probability that various types and sizes of releases occur. The PRA estimated how frequently each release
occurs and how much pathogen (or how many infected animals or people) is released. Releases could
occur via aerosol, via an infected worker or animal or via a contaminated worker.
For each release type, a different modeling approach was used. For releases that create an aerosol, we
used an atmospheric dispersion model to determine how many people or animals are exposed to what
dose of pathogen. Dose-response models were used to determine how many people or animals become
infected. For releases of contamination on the body or clothing of a laboratory worker, a stochastic,
Markov chain model was developed to determine how many people (if any) or how many animals (if any)
become contaminated after being touched by the initial contaminated worker. For events that caused the
infection of a laboratory worker, we used a stochastic model to determine if the worker violates the
various procedural and medical monitoring protocols to determine the probability of initiating an outbreak
in the community. As discussed further below, animal escapes were found to drive risk by escaping
containment features within the laboratory and infecting workers who can then leave the laboratory
instead of the animal escaping the laboratory entirely.4
Once a person was infected outside of the laboratory, we modeled the nascent local outbreak using a
branching process model, which captures the fact that small outbreaks can be extinguished by stochastic
factors and also public health measures, some of which may be unique to the communities around the
laboratories. Branching process models are stochastic models that calculate how many individuals every
contagious person infects in each generation. In our model, the probability of infecting a certain number
of new individuals is determined by the transmissibility of the pathogen (described by the parameter R),
and the variation in transmissibility between individuals (described by the parameter k) and modified by
public health control measures. We used this model to determine the probability that any given outbreak
would extinguish or grow beyond local control, given the properties of the pathogen, starting conditions
(how many of what type of people are infected) and control measures.
Once an outbreak escapes local control, it seeds outbreaks throughout the world. We modeled the
illnesses and deaths that occur in each region of the world using the HHS-BARDA Interactive Influenza
Model, an SEIR model that considers the effect of the young and the elderly in the ongoing spread of the
disease given contact rates between workplace, school and home populations. Although this study did not
attempt to evaluate the efficacy of public health response measures in detail, these measures must be
captured in our RA because some GoF phenotypes may vitiate some control measures more than others
(for example, the ability to overcome protective vaccination) and lead to a change in relative risk.
3 For example, see Vesely et al, Fault Tree Handbook (NUREG-0492), U.S. Nuclear Regulatory Commission, January 1981,
http://pbadupws.nrc.gov/docs/ML1007/ML100780465.pdf and Center for Chemical Process Safety, Guidelines for Hazard
Evaluation Procedures, April 2008.
4 The biosecurity section (Chapter 7) discusses an event that involves malicious actors stealing infected animals from a
laboratory. In this event, the malicious actor is assumed to be infected by carrying the infected animals, and the infection of
this person drives subsequent outbreak risk.

For pathogens that are transmissible amongst birds only, if a bird is infected outside of the laboratory, we
assumed that the outbreak escapes local control and that human health consequences are suffered based
on the consequences of past avian influenza outbreaks. This simple approach was taken because not
enough data is available to support more robust modeling of the interplay of wild birds, domestic birds
and humans for outbreaks caused by entirely novel, avian-transmissible pathogens that cannot be
transmitted amongst people. If an avian-origin strain is modified to be transmissible in mammals, it was
modeled as any other human-transmissible pathogen, as described above.
If a global epidemic is not triggered, consequences were tallied from the number of people infected in the
laboratory or in the smaller-scale outbreak in the locality. Comparing the risk posed by GoF research to
the risk posed by unmodified pathogens provides an understanding of which specific GoF experiments
may lead to a significant increase in biosafety risk.
Because the risk of a laboratory accident is proportional to the number of laboratory workers
manipulating dangerous strains of pathogens, we also characterized the proliferation potential of GoF
research in the US should the funding pause be lifted. We assessed the potential interest and capability to
perform GoF research in the US by an analysis of the scientific literature. We also examined funding
availability and the sufficiency of containment space to perform the work. Lastly, we identified three
cases of scientific discoveries in virology and traced their proliferation over the following years to
provide additional insight.
3.2 Biosecurity RA of malicious acts targeting a laboratory
In the risk assessment of malicious acts targeting a laboratory, also known as the semi-quantitative
biosecurity assessment, we compared the motivations and capabilities of a variety of malicious actors to
the defensive systems arrayed against them to prevent the malicious act. Should a malicious act lead to
the loss of containment, its consequences were modeled in the same manner as in the biosafety RA above.
No unclassified information describing the threats to research laboratories that store or study GoF
influenza, SARS, or MERS-CoV virus is available. Therefore, to identify the types of acts that may target
a GoF laboratory, our approach involved examining historical incidents involving life science laboratories
and hospitals, evaluating the motivations and capabilities of malicious actors, and determining if and how
existing security measures affect the likelihood of success of a malicious act. Plausible threats facing
laboratories that study or store GoF virus(s) were extrapolated from this assessment. From this
assessment, we can compare quantitatively how a malicious act would have different consequences if a
GoF laboratory was targeted instead of a laboratory studying only wild type pathogens.
To organize our biosecurity data collection effort, we developed a matrix of malicious actors, acts and
consequences. This matrix was reviewed by officials from the law enforcement and intelligence
communities to ensure that we captured all plausible combinations that could threaten biosecurity. We
then populated this matrix with data drawn from historical events that involved malicious acts in
laboratories in the US or overseas. This historical analysis provides an evidence-based method to
understand, in a qualitative way, the probability that an event would occur and the type of resources these
malicious actors bring to bear when targeting a laboratory.
To assess the capabilities of preventing malicious acts, we investigated the literature on legal authorities
and systems supporting biosecurity and analyzed these authorities and systems for gaps that could be
exploited by malicious actors. We also interviewed biosecurity stakeholders at institutions performing
relevant research to understand specific systems in place at these locations.

We then compared the data on the motives and capabilities of the malicious actors to the capabilities of
systems preventing their access to develop a series of qualitative scenarios that represent the “highest
risk” biosecurity events. The consequences of these events were modeled using the methodology laid out
in the biosafety RA as described above.
3.3 Biosecurity RA of GoF Information
In this assessment, we identified those GoF studies that, if published, would provide useful information
over what is already published in the scientific literature to a malicious actor seeking to create a biological
weapon. To perform this assessment, we first determined what is possible for a malicious actor to achieve
using unmodified agents so that we can identify how GoF pathogens could afford additional capabilities
to an adversary. We then characterized the state of the science regarding the enhancement of all traits
described in the NSABB GoF risk and benefit framework to understand to what degree methods already
exist in the literature that speak to the creation of modified strains of influenza viruses and coronaviruses
with phenotypes attractive to malicious actors. In this way, we identified GoF research that would provide
uniquely valuable information to a malicious actor for misuse over the body of dual-use research that
already exists. Also, we identified if dual-use information already in the literature requires a particularly
challenging technical approach in order to ascertain if an information risk could be suffered via the
publication of an easier experimental route to the same product. Lastly, we used open-source information
to understand if this unpublished dual-use information is actually desired by various malicious actors and
characterized the technical skill, sophistication and resources required for those actors to leverage this
information.
3.4 Benefit Assessment
The approach to the benefit assessment is founded on the concept that the benefits of scientific research
derive from applications of new scientific information or products to gaps in knowledge, public health,
medicine, and other societal issues. To that end, a multi-step process was used to identify the potential
benefits of GoF research.
1. A foundation for the analysis was established by independently:

a. characterizing the expected scientific information and products derived from GoF studies
of potential concern, and
b. identifying gaps in scientific knowledge about PPPs and gaps in public health and
medical capabilities related to the prevention and control of PPP outbreaks.
2. The scientific information/products derived from GoF research were mapped (“crosswalked”) to
the gaps in scientific knowledge, public health, and medicine,
3. Alternate experimental approaches and/or other scientific or technical innovations (“alt-GoF”

approaches) that could address the same gaps were identified,
4. The barriers to the realization of GoF and alt-GoF benefits were evaluated,
5. The unique benefits of GoF research were identified by comparatively analyzing the benefits
afforded by GoF research versus alternative approaches, in light of the barriers to the realization
of each approach,
6. The potential for the unique benefits of GoF research to be globalized was assessed, and

7. Benefits to the production of influenza vaccines were quantitatively evaluated.
This analysis of GoF benefits was guided by the benefits of GoF research proposed by infectious disease
researchers and other GoF stakeholders during public meetings about GoF research and through
perspectives and research articles published in scientific journals. This data set was reinforced and
expanded upon via interviews with stakeholders involved in conducting scientific research and translating
it into practice for the benefit of public health or medicine. Importantly, this analsyis leveraged the
evaluation of public health systems to understand the process by which the immediate applications of
GoF research ultimately reduce human morbidity and mortality caused by influenza viruses and
coronaviruses. Taken together, this analysis resulted in the identification of GoF research outputs with
validated applications to scientific knowledge, public health, and medicine as well as an understanding of
their downstream benefits to the health of human populations.
GoF studies comprise a subset of all research activities involving PPPs, and some alternative
experimental approaches or other scientific/technical innovations (hereafter referred to as “alt-GoF”
approaches) may pose less risk than GoF studies but yield the same or similar benefits. The potential
benefits of alt-GoF studies were identified through the same process as for GoF studies: a crosswalk of
the research outputs of alt-GoF studies to gaps in scientific knowledge, public health, and medicine
related to PPPs. Importantly, alt-GoF approaches also include those scientific and technical innovations
that address the same public health gaps that GoF can address but through a completely different
mechanism. To complement the analysis of the net risks associated with the conduct of GoF research
relative to research involving wild type pathogens, the benefit assessment highlights those types of GoF
studies that may provide unique benefits to scientific knowledge, public health, and medicine relative to
alternative approaches.
One of the most challenging aspects of weighing the risks and benefits of GoF research is that there is a
temporal mismatch between the risks and the benefits of the research – the risks are assumed at the time
the research is conducted, while the benefits to public health and medicine may accrue in the future. To
enable the comparison of risks and benefits, the benefit assessment provides data on the likelihood that
the potential benefits of GoF research will be realized by describing the barriers to the realization of the
benefits. To assess the barriers to benefits in public health and medicine, the gap analysis of public health
and medical capabilities related to the prevention and control of PPP outbreaks was leveraged. In
addition, uncertainties in the meaning of the scientific outcomes of GoF studies, which may represent
scientific barriers to the application of GoF research, were explored through analysis of the scientific
literature and interviews with infectious disease researchers. Together, this method enables the
identification of the type of resources required to overcome or circumvent each barrier, including
advancements in scientific knowledge, improvements to public health infrastructure, and other factors,
which serves as a proxy for the likelihood and timing of the realization of the benefits.
Whether risks and benefits are equally distributed across populations is also an important consideration in
any risk-benefit comparison. To inform NSABB’s deliberations on this issue, the benefit assessment
qualitatively assessed the globalization potential of the identified GoF benefits, through analysis of
historical case studies examining the globalization of similar benefits and through review of relevant USG
policies on resource and information sharing. Benefits related to the production of influenza vaccines are
amenable to quantitative analysis. This analysis leveraged models developed for the biosafety RA above
to parametrically explore how changes in the availability of influenza vaccines can mitigate morbidity or
mortality during seasonal flu epidemics and flu pandemics.

4 Background Information on Influenza Viruses, the
Coronaviruses and GoF Research
Influenza Viruses 17
4.1.1 Biology of Influenza 17
4.1.2 Influenza Epidemiology 22
4.1.3 Asymptomatic Infections 24
4.1.4 Symptomatic Infections 25
4.1.5 Mortality 25
The SARS- and MERS-coronaviruses 26

4.2.1 Biology of the coronaviruses 26
4.2.2 Genome Structure of the coronaviruses 26
4.2.3 Diversity of the coronaviruses 28
4.2.4 Host Range of the coronaviruses 29
4.2.5 SARS-CoV Epidemiology 30
4.2.6 MERS-CoV Epidemiology 32
An overview of GoF Research 34

4.3.1 Coronaviruses 35
4.3.2 Influenza viruses 37

4.1 Influenza Viruses
Throughout this report, the terms pandemic, seasonal and avian are used. Seasonal influenza viruses
include the strains of the H1N1 and H3N2 subtypes that cause morbidity on an annual basis. Pandemic
influenza viruses include the 1918 H1N1, 1957 H2N2, 1968 H3N2 and 2009 H1N1 strains, which spread
rapidly at least partially because the population had very little immunity. The 2009 H1N1 strain continues
to circulate seasonally since its emergence, and is properly classified as both a pandemic strain and a
seasonal strain (also the morbidity and mortality caused by this strain is more similar to seasonal strains
than any of the previous pandemic strains). Similarly, today’s seasonal H1N1 strains are descendants of
the 1918 H1N1 strain and the seasonal H3N2 strains are descendants of the 1968 strain, so the distinction
between pandemic and seasonal strains is one of timing (today’s novel pandemic strain is tomorrow’s
seasonal strain).
Avian influenza strains are strains of influenza that are transmissible only amongst animals other than
humans, especially birds. If an avian strain is modified to become transmissible in humans, we still call
this an avian strain because of the characteristics of its wild type parents. Generally, this report is
concerned with the highly pathogenic strains of the H5 and H7 subtypes.
4.1.1 Biology of Influenza
4.1.1.1 Overview
Influenza is a single-stranded, negative-sense RNA virus of the Orthomyxoviridae family. There are three
types of influenza viruses—A, B, and C—that have a common genetic ancestry, but distinct genetic
characteristics. Influenza type A can infect a variety of animal hosts and is further divided into subgroups
based on its surface proteins. Type B viruses have a more limited host range with limited variation;
influenza C causes only mild symptoms in humans and does not contribute to outbreaks.5
4.1.1.2 Virus Structure
The influenza genome is divided into eight RNA segments that encode viral proteins essential to the
functionality of the virus. Each is folded into a rod-shaped, double-helical ribonucleoprotein complex
(RNP).
5 Centers for Disease Control and Prevention. Types of Influenza Viruses. http://www.cdc.gov/flu/about/viruses/types.htm.
Last Update 2014. Accessed May 2015.

Figure 4.1. An influenza virion displaying the segmented RNA genome and encoded proteins, as reproduced
from Horimoto et al.6
The RNP contains viral RNA encapsulated by nucleoprotein (NP) and bound to a trimeric RNA
polymerase, comprised of one polymerase acidic (PA) and two polymerase basic (PB1, PB2) subunits.
The RNP is responsible for directing RNA replication, transcription, and transport as well as genome
reassortment and packaging. Nuclear export proteins (NEP) also facilitate intracellular transport.
Matrix proteins, M1, surround the RNPs and NEPs. The lipid bilayer envelope encloses the virion with
hemagglutinin (HA), neuraminidase (NA), and matrix ion channel (M2) proteins embedded into its
membrane. HA glycoproteins facilitate virus binding and entry whereas NA proteins promote virus
budding. HA recognizes the sialic acid moieties on host cells and ensure proper binding in preparation for
endocytosis. NA proteins possess sialidase activity to release newly replicated virus from and prevent
virus aggregation on the host cell. M2 ion channel proteins are vital for pH regulation during viral
replication.7,8,9
4.1.1.3 Antigenic Variation
All influenza viruses are classified by type and strain. Influenza A viruses are also classified by their HA
and NA subtype—e.g. H1N1, H3N2, H5N1. Influenza is a relatively simple RNA virus, yet is able to
continuously elude host immune systems through antigenic drift. Influenza’s RNA polymerase is prone to
replication errors, resulting in frequent point mutations in antibody binding sites on HA and NA proteins.
These amino acid changes have the potential to affect the conformation of surface proteins, and hence, the
binding of host antibodies. Although these mutations are minor and random, accumulation over time can
lead to a new strain of virus that is no longer neutralized by the host immune system, even after
vaccination or prior infection.
Antigenic drift occurs in both influenza A and B. Influenza A viruses, however, can also evolve through
a much more abrupt process referred to as antigenic shift, which is the resilt of genomic reassortment. The
segmented feature of the virus genome enables the entire HA or NA segment to be replaced with a new
6 Horimoto T, Kawaoka Y (2005) Influenza: lessons from past pandemics, warnings from current incidents. Nature reviews
Microbiology 3: 591-600
7 Bouvier NM, Palese P (2008b) The biology of influenza viruses. Vaccine 26, Supplement 4: D49-D53
8 Shaw ML et al (2008) Cellular proteins in influenza virus particles. PLoS Pathog 4: e1000085
9 Tsai KN, Chen GW (2011) Influenza genome diversity and evolution. Microbes Infect 13: 479-488

segment from a different influenza virus. By altering the surface proteins, the infectivity of the virus is
also altered and a new phenotypic subtype is formed. The HA protein is more likely to be reassorted, but
both HA and NA subtypes have variable antigenicity.
Genomic reassortment occurs as a result of co-circulation of different subtypes of influenza A and co-
infection of a host. When a host is infected with two influenza strains, viral replication in the nucleus may
cause mixing of genetic material. Influenza is prone to genetic mixing due to its multiple-stranded
genome (that is, a single newly budded virus particle could package RNA strands from two different
parental strains). This mixing can result in a significant change of the antigenic properties of the virus,
termed antigenic shift, and could generate a virus to which hosts have no existing immunity and,
therefore, is the source of pandemic outbreaks. After a virus undergoes antigenic shift, it continues to
experience antigenic drift. Both antigenic drift and antigenic shift are responsible for influenza’s
evolution and survival.10,11
Figure 4.2. Antigenic drift and antigenic shift in influenza A virus, as reproduced from the WHO
Collaborating Centre for Reference and Research on Influenza.12
4.1.1.4 Host Range
Influenza types, subtypes, and strains have a distinct and sometimes overlapping set of host organisms
that they can effectively infect, called their host range or host tropism. For influenza viruses, sialic acid
receptor specificity, temperature, and pH at the site of infection are the main determinants of host range.
Receptor specificity largely determines the host tropism of a virus. HA proteins bind to host glycosylated
receptors with sialic acid moieties; different HA subtypes have preferences for different bond structures.
HA in avian viruses binds only to the α-2,3 isoform whereas in human-adapted viruses, the α-2,6 linkage
is preferred (Figure 4.3). Either HA type, however, will bind in swine because the species possesses cells
with both sialic acid moieties. For this reason, swine are considered “mixing vessels” that provide an
10 Carrat F, Flahault A (2007) Influenza vaccine: the challenge of antigenic drift. Vaccine 25: 6852-6862
11 Bouvier NM, Palese P (2008a) The biology of influenza viruses. Ibid. 26 Suppl 4: D49-53
12 WHO Collaborating Centre for Reference and Research on Influenza. About Influenza.
http://www.influenzacentre.org/aboutinfluenza.htm. Last Update Accessed October 2015.

opportunity for reassortment. Reassortment and evolution in intermediate hosts allow for emergence of
new virus types. Figure 4.4 below shows the species adapted to different HA and NA subtypes.13,14
Figure 4.3. Chemical structures of α-2,3- and α-2,6-linked glycans, with the terminal sialic acid and galactose,
for binding to influenza viruses, as reproduced from Xu et al.15
13 Xu R et al (2010) Structure, receptor binding, and antigenicity of influenza virus hemagglutinins from the 1957 H2N2
pandemic. J Virol 84: 1715-1721
14 Medina RA, Garcia-Sastre A (2011) Influenza A viruses: new research developments. Nature reviews Microbiology 9: 590-
603
15 Xu R et al (2010) Structure, receptor binding, and antigenicity of influenza virus hemagglutinins from the 1957 H2N2
pandemic. J Virol 84: 1715-1721

Figure 4.4. Venn diagram of species infected by influenza types and subtypes as reported by the CDC and
WHO, adapted from McCandless et al. Size of text represents human fatality rates and lighter text shows the
virus rarely infects humans.16
For effective transmission to another host, the virus must be able to efficiently replicate in the temperature
of the new host and the site of infection in each host has a distinct temperature range. In birds, infection
occurs in the gastrointestinal tract at around 40 degrees Celsius. In swine, influenza targets the respiratory
tract at approximately 39 degrees Celsius. The upper respiratory tract in humans is typically around 33
degrees Celsius whereas the lower respiratory tract reaches 37 degrees Celsius. The lower respiratory
tract also hasα-2,3 moieties, which can be bound by avian strains. The elevated temperature, in
combination with avian compatible viral receptors, presents the opportunity for a non-human-adapted
virus to infect a human. Although rare, this is a source of unexpected species crossover, creating a variant
influenza strain. The host range cycle can be seen in Figure 4.5.17,18,19
16 McCandless D, Hollowood E. Influ-Venn-Za. Who can catch which flu?

http://www.informationisbeautiful.net/visualizations/which-flu-virus/. Last Update April 2013. Accessed October 2015.
17 Medina RA, Garcia-Sastre A (2011) Influenza A viruses: new research developments. Nature reviews Microbiology 9: 590-
603
18 Causey D, Edwards SV (2008) Ecology of avian influenza virus in birds. J Infect Dis 197 Suppl 1: S29-33
19 Cheng VC et al (2012) Two years after pandemic influenza A/2009/H1N1: what have we learned? Clin Microbiol Rev 25:
223-263

Figure 4.5. The host range of an influenza virus subtype is dependent on sialic acid receptor specificity and
infectivity temperature of the host, as adapted from Cheng et al.20
HA glycoproteins facilitate viral infection of a host cell through pH-induced membrane fusion and some
level of host tropism is determined by the pH of the infection site in various hosts. Change in pH may
render the virus ineffective at transferring its genome into the host cell by causing the virus to release its
genome at less proximity to the nucleus or once lysosomes have matured to degrade the genome. Either
will inhibit viral infection and replication. Viral adaption to a new host species requires a pH shift for
effective membrane fusion.21
Currently only H1, H2, and H3 subtypes can easily cause human-to-human transmission. The emergence
of a new subtype capable of human to human transmission is expected to cause a major pandemic because
the population will have no pre-existing immunity.
4.1.2 Influenza Epidemiology22
Influenza is an acute viral infection characterized by the rapid onset of disease and brief symptomatic
period. The virus can cause mild to severe respiratory illness in human hosts; some infections cause minor
respiratory symptoms while others result in hospitalization and occasionally, death. Unresolved cases are
usually associated with other chronic conditions and can develop into additional complications such as
pneumonia and bronchitis.
20 Ibid.
21 Mair CM et al (2014) Receptor binding and pH stability — How influenza A virus hemagglutinin affects host-specific virus
infection. Biochimica et Biophysica Acta (BBA) - Biomembranes 1838: 1153-1168
22 Much of the data and sources discussed below were drawn from a previous study completed for the Defense Threat
Reduction Agency by Gryphon Scientific called Influenza Modeling Parameters.

4.1.2.1 Incubation Period
The incubation period is the time between when an individual is exposed to a pathogen and when the first
symptom manifests. During the incubation period, most infected individuals cannot transmit the infection
to others; therefore, longer incubation periods equate to a slower outbreak development. Incubation
periods vary for seasonal, pandemic, and severe pandemic influenza.
4.1.2.1.1 Seasonal Influenza
Several papers were identified that describe the incubation periods observed in seasonal influenza
infections. The literature suggests an incubation period duration ranging from one day to seven days
(Supplemental Information Table 1). The most common incubation period found within the literature was
two days with a mean incubation period of 63 hours or 2.6 days. 23,24,25,26,27,28,29
4.1.2.1.2 Pandemic Influenza
Incubation periods for pandemic influenza are reported to be slightly longer than those seen in seasonal
influenza. Since there is little to no data on the incubation period of other pandemic strains, data from the
2009 H1N1 outbreak were evaluated. Four sources from the 2009 H1N1 pandemic reported data on the
length of incubation periods.
The H1N1 data suggest a range of incubation periods similar to the range seen in seasonal influenza.
However, these data suggest a mean incubation period of 4.2 days and median of 4.1 days instead of the
two day incubation period most commonly seen in seasonal influenza.30,31,32,33, 34
4.1.2.2 Infectious Period
The infectious period is the disease stage when an infected individual can transmit their disease to others.
Currently, however, there is no definitive way to determine when an individual infected with influenza
virus is contagious. The most widely accepted method of determining contagiousness is by measuring
viral shedding. Under this method, an individual is deemed infectious when they begin shedding virus and
23 Alford RH et al (1966) Human influenza resulting from aerosol inhalation. Experimental Biology and Medicine 122: 800-
804
24 Burnet F, Foley M (1940) The Results of Intranasal Inoculation of Modified and Unmodified Influenza Virus Strains in
Human Volunteers. Medical Journal of Australia 2: 655-659
25 Couch RB et al (1971) Correlated studies of a recombinant influenza-virus vaccine. III. Protection against experimental
influenza in man. Journal of Infectious Diseases 124: 473-480
26 Macdonald P, Lyth JC (1918) INCUBATION PERIOD OF INFLUENZA. Br Med J 2: 488
27 Moser MR et al (1979) An outbreak of influenza aboard a commercial airliner. American journal of epidemiology 110: 1-6
28 Armstrong C, Hopkins R (1921) An epidemiological study of the 1920 epidemic of influenza in an isolated rural
community. Public Health Reports (1896-1970): 1671-1702
29 Lessler J et al (2009) Incubation periods of acute respiratory viral infections: a systematic review. The Lancet infectious
diseases 9: 291-300
30 Cao B et al (2009) Clinical features of the initial cases of 2009 pandemic influenza A (H1N1) virus infection in China. New
England Journal of Medicine 361: 2507-2517
31 Li H, Wang SX (2010) Clinical features of 2009 pandemic influenza A (H1N1) virus infection in chronic hemodialysis
patients. Blood Purif 30: 172-177
32 Tuite AR et al (2010) Estimated epidemiologic parameters and morbidity associated with pandemic H1N1 influenza.
Canadian Medical Association Journal 182: 131-136
33 Wang C et al (2012) Epidemiological and clinical characteristics of the outbreak of 2009 pandemic influenza A (H1N1) at a
middle school in Luoyang, China. Public Health 126: 289-294
34 Ghani A et al (2009) The Early Transmission Dynamics of H1N1pdm Influenza in the United Kingdom. PLoS currents 1:
RRN1130

stops being infectious when the viral shedding ends. The infectious period of seasonal and pandemic
influenza is seemingly the same.
4.1.2.2.1 Seasonal and Pandemic Influenza
Data on when individuals infected with influenza stop shedding virus are extremely limited. The few
availably papers typically assume that viral shedding begins at the onset of symptoms, and report only the
average time after symptom onset when viral shedding ceases at 5.9 days.35,36 Only Doyle et al. reported
the distribution of viral shedding durations in addition to average duration.37 In this study individuals
were experimentally infected with influenza H1N1 virus (a pandemic strain) and monitored daily for viral
shedding. All infected individuals shed virus for a minimum of three days after onset of symptoms, and a
small percentage of individuals shed for eight or more days (see Supplemental Information on influenza
disease course). However, more than 50% of those infected shed virus for six or seven days. The study
also provided evidence that viral shedding occurred before symptoms were displayed, which would
increase the total time of shedding. No other sources were available on the duration of viral shedding for
influenza or to confirm viral shedding before symptoms.
4.1.3 Asymptomatic Infections
A small portion of individuals infected with influenza virus never get clinically ill. These asymptomatic
individuals are infected with influenza, shed virus, and therefore have the potential to transmit to others,
but never develop symptoms.
4.1.3.1 Seasonal Influenza
Several studies examined the percent of asymptomatic seasonal influenza infections. Data from three
papers, Lau et al., Loeb et al., and Suess et al., were included in our analysis (Supplemental Information
on influenza disease course). These three studies all used the same method for defining an asymptomatic
infection.38 Individuals were considered to be asymptomatic if they were actively shedding influenza
virus but were not experiencing any upper respiratory infection symptoms. Individuals that were exposed
to influenza through a close contact (usually a family member) were monitored for influenza viral
shedding to determine if an infection had occurred. Infected individuals were then monitored to determine
whether or not symptoms developed. The three studies suggest that 13% of individuals infected with
seasonal influenza virus experience an asymptomatic infection. 39,40,41
35 Carrat F et al (2008) Time lines of infection and disease in human influenza: a review of volunteer challenge studies.
American journal of epidemiology 167: 775-785
36 Lau LL et al (2010) Viral shedding and clinical illness in naturally acquired influenza virus infections. Journal of Infectious
Diseases 201: 1509-1516
37 Doyle WJ et al (1998) Effect of rimantadine treatment on clinical manifestations and otologic complications in adults
experimentally infected with influenza A (H1N1) virus. J Infect Dis 177: 1260-1265
38 A 2008 paper by Carrat et al. also reviewed this topic; however, it did not explain how “asymptomatic” infections were
defined and was therefore excluded from our analysis. Carrat F et al (2008) Time lines of infection and disease in human
influenza: a review of volunteer challenge studies. American journal of epidemiology 167: 775-785
39 Lau LL et al (2010) Viral shedding and clinical illness in naturally acquired influenza virus infections. Journal of Infectious
Diseases 201: 1509-1516
40 Loeb M et al (2012) Longitudinal study of influenza molecular viral shedding in Hutterite communities. Journal of
Infectious Diseases 206: 1078-1084
41 Suess T et al (2012) Comparison of shedding characteristics of seasonal influenza virus (sub) types and influenza A (H1N1)
pdm09; Germany, 2007–2011. PloS one 7: e51653

4.1.3.2 Pandemic Influenza
Only one paper was identified that examined asymptomatic pandemic influenza infections. During the
2009 H1N1 pandemic, Papenburg et al. used the same techniques described by Lau et al., Loeb et al., and
Suess et al. in which asymptomatic individuals that shared a household with symptomatic individuals
were monitored for viral shedding.42 Papenburg et al. found 9.4% of individuals that shed H1N1 influenza
virus remained symptom free.
4.1.4 Symptomatic Infections
An influenza diagnosis encompasses a variety of symptoms that can manifest in different combinations
within each individual. Many symptoms are shared by seasonal and pandemic influenza, but some are
only produced by more severe pandemic infections. Symptoms associated with seasonal influenza include
chills, cough, diarrhea, fatigue, fever, headaches, myalgia, nasal congestion, rhinorrhea, sore throat, and
vomiting. Additionally, pandemic influenza can also cause abdominal pain, bronchospasms, chest pain,
confusion, conjunctivitis, loss of appetite, nosebleeds, and seizures. Not every individual will experience
all of the symptoms for each disease.
4.1.4.1 Seasonal and Pandemic Influenza
The prevalence of some influenza symptoms are also age dependent.43 For example, children with
seasonal influenza are significantly more likely to experience vomiting than are those who are 60 years
and older. The prevalence of each influenza symptom varies between children, adults, and the elderly
during seasonal and pandemic outbreaks (Supplemental Information on influenza disease course).
Data on the prevalence of symptoms were obtained from observational influenza studies and from the
control subjects of anti-influenza neuraminidase inhibitor clinical trials. These studies recorded the
number and/or percentage of people experiencing an influenza infection and the specific symptoms they
developed. No data on pandemic influenza in the elderly was identified. 44
4.1.5 Mortality
Almost all infected patients will fully recover. A small portion of illnesses, however, will end in death.
Thompson et al. in 2003 analyzed and abstracted seasonal influenza data compiled by the National Center
for Health Statistics (NCHS) from 1990-1998. These data were then used to estimate the rate of
influenza-associated deaths by age groups (Supplemental Information on influenza disease course). The
percent mortality of infected individuals ranges from 0.0002% for those between five and 49 years old to
0.02% for those older than 64.45
42 Papenburg J et al (2010) Household transmission of the 2009 pandemic A/H1N1 influenza virus: elevated laboratory-
confirmed secondary attack rates and evidence of asymptomatic infections. Clinical Infectious Diseases 51: 1033-1041
43 Centers for Disease Control and Prevention. Flu Symptoms & Severity. Retrieved from
http://www.cdc.gov/flu/about/disease/symptoms.htm. Last Update September 2014. Accessed May 2014.
44 Cox NJ, Subbarao K (1999) Influenza. The Lancet 354: 1277-1282
45 Thompson WW et al (2003) Mortality associated with influenza and respiratory syncytial virus in the United States. Jama
289: 179-186

4.1.5.2 Pandemic Influenza
The mortality rate of pandemic influenza is both difficult to estimate or predict due to the limited number
of past pandemic outbreaks. It is estimated that approximately 500 million people were infected with the
1918 Spanish flu and 50 to 100 million people perished as a result of infection.46 During the 2009 H1N1
pandemic, there were anywhere from 43 to 89 million cases of influenza with resultant 9,000 to 18,000
deaths.47 The percentage of influenza infections that resulted in mortality was approximately 5% during
the 1918 pandemic and less than 0.005% during the 2009 pandemic (Supplemental Information on
influenza disease course).
4.2 The SARS- and MERS-coronaviruses
Throughout this report, our use of the term “coronaviruses” or “CoVs” refers specifically to SARS-CoV,
MERS-CoV, and SARS/MERS-like bat CoVs such as HKU4 and HKU5. Note, the four human
coronaviruses that cause mild to moderate respiratory illnesses such as the common cold or croup
(coronaviruses HKU1, OC43, 229E, and NL63) were not evaluated because these are not considered in
the NSABB GoF Framework.
4.2.1 Biology of the coronaviruses
Severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) are caused by
SARS-associated (SARS-CoV) and MERS-associated coronavirus (MERS-CoV), respectively.
Coronaviruses are positive sense, single-stranded RNA viruses. They are the largest of all RNA viruses,
comprised of approximately 30 thousand nucleotides. Due to the length of its genome, coronaviruses can
be less dependent on cellular proteins than other RNA viruses, enabling easier cross-species transmission.
Three groups of coronaviruses have been identified, all with distinct genetic and serological identities.
While they are both beta-coronaviruses, MERS-CoV is from lineage B while SARS-CoV belongs to
lineage C.48 Aside from SARS-CoV and MERS-CoV, coronaviruses that can infect humans cause
common colds, lower respiratory tract infections, and diarrhea.49,50
4.2.2 Genome Structure of the coronaviruses
The long SARS-CoV genome is broken into five major open reading frames (ORF), which are sections of
nucleotides responsible for coding a peptide (Figure 4.1). Beginning at the five prime end, the first two
ORFs, 1a and 1b, comprise two-thirds of the genome and encode the viral replicase genes, which encode
proteins that are responsible for viral genome replication in the host cell. Further down the genome,
ORFs encode genes for the structural proteins of SARS-CoV: spike (S), envelope (E), membrane (M),
and nucleocapsid (N). These characterized ORFs are interspaced between several other ORFs that encode
46 Taubenberger JK, Morens DM (2006) 1918 Influenza: the mother of all pandemics. Emerging infectious diseases 12: 15-22
47 National Institute of Allergy and Infectious Diseases. (2011) Pandemic Flu History. Department of Health & Human
Services, Washington, DC.
48 Hilgenfeld R, Peiris M (2013) From SARS to MERS: 10 years of research on highly pathogenic human coronaviruses.
Antiviral Res 100: 286-295
49 Satija N, Lal SK (2007) The molecular biology of SARS coronavirus. Ann NY Acad Sci 1102: 26-38
50 Li W et al (2006) Animal origins of the severe acute respiratory syndrome coronavirus: insight from ACE2-S-protein
interactions. J Virol 80: 4211-4219

accessory genes. While the exact role of accessory genes is unknown, they are believed to contribute to
viral pathogenesis and not replication.51,52
MERS-CoV has a similar genome structure, including viral replicase genes and structural proteins, as
SARS-CoV.53
4.2.2.1 Structural Proteins and Particle Structure
The M glycoprotein is responsible for virus assembly. M proteins are the most abundant transmembrane
protein in the viral envelope, where they interact with N proteins. N proteins self-associate to helically
encapsidate the viral RNA and form the ribonucleoprotein complex (RNP). Together with M proteins,
these N proteins mediate incorporation of the genome into budding virions for release. N proteins are
highly immunogenic and their interaction with host cell proteins establishes pathogenicity. Envelope (E)
proteins are a hydrophobic integral membrane proteins that serve as viroporins, which form ion channels
in the envelope and therefore, play a central role in virus morphogenesis and assembly. E proteins are also
credited with preserving the membrane’s curvature for particle stability and infectivity.
Lastly, spike proteins are transmembrane fusion proteins responsible for effective viral entry into host
cells. The N-terminal domain (S1) facilitates target receptor binding while the C-terminal domain (S2)
ensures proper viral fusion. The S1 domain differs between coronavirus types and is largely responsible
for host range. 54 Activated spike proteins induce the host immune response, including antibody
neutralization and are the major antigenic determinants of MERS-CoV and SARS-CoV. 55,56,57,58
The virion contains one copy of the viral genome encapsidated by N proteins in an RNP (Figure 4.6). A
viral envelope surrounds the virion with structural proteins S, E, and M embedded in its membrane.
Coronavirus has a crown-like appearance due to the protruding club-shaped spike proteins.59
51 Kopecky-Bromberg SA et al (2007) Severe acute respiratory syndrome coronavirus open reading frame (ORF) 3b, ORF 6,
and nucleocapsid proteins function as interferon antagonists. Ibid. 81: 548-557
53 Coleman CM, Frieman MB (2013) Emergence of the Middle East respiratory syndrome coronavirus. PLoS Pathog 9:
e1003595
54 Li F (2015) Receptor recognition mechanisms of coronaviruses: a decade of structural studies. J Virol 89: 1954-1964
55 Siu YL et al (2008) The M, E, and N structural proteins of the severe acute respiratory syndrome coronavirus are required
for efficient assembly, trafficking, and release of virus-like particles. Ibid. 82: 11318-11330
56 Tan YJ et al (2005) Characterization of viral proteins encoded by the SARS-coronavirus genome. Antiviral Res 65: 69-78
interactions. J Virol 80: 4211-4219
59 Ibid.

Figure 4.6. The SARS coronavirus (SARS-CoV) genome and virion structure, as reproduced from Perlman et
al. 60 MERS-CoV possesses a similar genome and virion structure.61
4.2.3 Diversity of the coronaviruses
Coronaviruses evolve rapidly, similar to all RNA viruses because polymerase infidelity results in amino
acid mutations that alter transcription and potentially translation. Although these genetic mutations are
minor and random, accumulation over time can lead to a new strain of virus. Some believe that the
unusually large coronavirus genome leads to more mutations. Other studies have shown, however, that
the genome encodes additional RNA processing and editing enzymes to correct for polymerase
errors.62,63,64
The coronavirus genome is also prone to homologous recombination. Recombination allows for genetic
exchange between different virus strains during coinfection. The natural process facilitates cross-species
transmission and the generation of new coronavirus species. 65,66,67
60 Perlman S, Dandekar AA (2005) Immunopathogenesis of coronavirus infections: implications for SARS. Nature reviews
Immunology 5: 917-927
61 Coleman CM, Frieman MB (2013) Emergence of the Middle East respiratory syndrome coronavirus. PLoS Pathog 9:
e1003595
62 Graham RL, Baric RS (2010) Recombination, reservoirs, and the modular spike: mechanisms of coronavirus cross-species
transmission. Journal of virology 84: 3134-3146
interactions. Ibid. 80: 4211-4219
64 Dudas G, Rambaut A (2015) MERS-CoV recombination: implications about the reservoir and potential for adaptation.
bioRxiv
67 Dudas G, Rambaut A (2015) MERS-CoV recombination: implications about the reservoir and potential for adaptation.
bioRxiv

Recombination can affect all viral proteins, but variation in the spike protein has a considerable effect on
the virus due to its role in viral entry and host range. Changes in virulence, species transmission patterns,
and host range are often a result of spike protein recombination.68,69
4.2.4 Host Range of the coronaviruses
4.2.4.1 SARS-CoV
Before infecting humans, SARS-like CoVs infected an array of other animal species, including bats, palm
civets, monkeys, domestic cats, raccoons, and ferrets. Bats are the virus’s natural reservoir, however,
palm civets are credited as the amplifying host that transmitted SARS-CoV to humans.70
Host specificity of SARS-CoV is heavily influenced by receptor recognition and hence, its spike protein.
Viral sequencing suggests that the spike protein experienced heavy positive selection at the onset of the
SARS outbreak. 71 Clinical data supports this premise, as SARS-CoV became increasingly pathogenic and
transmissible among humans as the epidemic progressed; virus evolution through mutations to the spike
protein were the likely cause. 72,73
SARS-CoV entry is mediated by angiotensin I converting enzyme 2 (ACE2), the host cell receptor
(Figure 4.2). Ordinarily ACE2 regulates host blood pressure. Host susceptibility to the SARS coronavirus
is dependent on the binding affinity between the virus and the host-specific ACE2. Only two residue-
altering mutations in the ACE2 gene were necessary to overcome the species barrier between palm civets
and humans leading to sustained human infection. 74,75
ACE2 is primarily found on ciliated cells in the lung epithelia, which explains the tropism of SARS-CoV
to the lungs and the resultant respiratory illness. These receptors have also been detected in the heart,
colon, and kidneys.76,77 The absence of this receptor in muscle, blood or skin cells suggest that there is
very little risk of infection if SARS-CoV is introduced in a cut.
4.2.4.2 MERS-CoV
The MERS-CoV spike protein, just as in SARS-CoV, largely determines the host range of the virus.
Dipeptidyl peptidase 4 (DPP4), also known as CD26, was recently identified as the host receptor for viral
entry. DPP4 is a widely expressed cellular protease tasked with assisting immune responses, glucose
metabolism, and apoptosis. The glycoprotein is found on many cellular surfaces, including the kidneys,
70 Li F (2013) Receptor recognition and cross-species infections of SARS coronavirus. Antiviral Res 100: 246-254
71 Sheahan T et al (2008) Mechanisms of zoonotic severe acute respiratory syndrome coronavirus host range expansion in
human airway epithelium. Journal of virology 82: 2274-2285
72 ibid.

lungs, small intestines, and liver, which accounts for the virus’s ability to cause systemic infection and
shock.78,79
MERS-CoV infects a larger range of animals than does SARS-CoV. Transmission has occurred through
close contact between humans and animals, most likely dromedary camels or bats.80 The MERS
coronavirus can also infect primates, horses, and goats, but is ineffective in smaller mammals such as
ferrets, hamsters, and mice.81 Host susceptibility to MERS is dependent on the binding affinity between
the virus and the host-specific DPP4. Differences have been detected in DPP4 glycoproteins among
mammals that may alter such affinity.82
There are no similarities between the structure or sequence of DPP4 and ACE2, the host receptor for
SARS-CoV, which explains the distinct host ranges among the two coronaviruses. Further research
suggests that differences in expression levels and locations of the receptors may account for the viruses’
difference pathogenesis.83
4.2.5 SARS-CoV Epidemiology
SARS is an acute viral respiratory illness that develops into severe pneumonia. It is a contagious and
virulent disease. Without treatment, the pneumonia may lead to respiratory failure and death.84
4.2.5.1 Incubation period
The incubation period is the time between when an individual is exposed to a pathogen and when the first
symptom manifests. During the incubation period of SARS and MERS, infected individuals probably
cannot transmit the infection to others; therefore, longer incubation periods equate to a slower outbreak
development.
The incubation period of SARS is was found to vary significantly between patients and during the 2003
pandemic, between countries. According to the World Health Organization (WHO), most countries
experienced a median incubation period of 4-5 days and mean of 4-6 days with a minimum of 1 day and
maximum of 14 days reported.85 The primary literature is rich with studies on the incubation periods of
SARS cases (Supplemental Information on CoV disease course). The literature suggests a mean
78 Abdel-Moneim AS (2014) Middle East respiratory syndrome coronavirus (MERS-CoV): evidence and speculations. Arch
Virol 159: 1575-1584
79 Peck KM et al (2014) Coronavirus Host Range Expansion and Middle East Respiratory Syndrome Coronavirus Emergence:
Biochemical Mechanisms and Evolutionary Perspectives. Annual Review of Virology
80 Penttinen PM et al (2013) Taking stock of the first 133 MERS coronavirus cases globally--Is the epidemic changing? Euro
surveillance : bulletin Europeen sur les maladies transmissibles = European communicable disease bulletin 18
81 Peck KM et al (2014) Coronavirus Host Range Expansion and Middle East Respiratory Syndrome Coronavirus Emergence:
Biochemical Mechanisms and Evolutionary Perspectives. Annual Review of Virology
82 Wang N et al (2013) Structure of MERS-CoV spike receptor-binding domain complexed with human receptor DPP4. Cell
Res 23: 986-993
83 Ibid.
84 The Centers for Disease Control and Prevention (2004) Basic Information about SARS Fact Sheet
85 The World Health Organization, Severe Acute Respiratory Syndrome (SARS) Epidemiology Working Group (2003a)
Consensus document on the epidemiology of severe acute respiratory syndrome (SARS).

incubation period of 5.18 days and median of 4 days. 86,87,88,89,90 Donnelly found that 95% of patients
experience onset of symptoms within 14.22 days.91 The literature findings generally support the published
reports from the WHO but presented a range of 1-18 days, capturing the variability of the SARS
incubation period. No definitive explanations exist for the cause of the distribution of incubation period is
so widespread, but difficulty identifying exposure, varying infectious doses, and multiple exposures are
possible causes. Route of transmission may also affect the incubation period, but it is unclear why or how.
92,93,94,95,96
4.2.5.2 Infectious period
The infectious period is the disease stage when an infected individual can transmit the disease to others.
The most widely accepted method of determining contagiousness is measuring viral shedding. Under this
method, an individual is deemed infectious when they begin shedding virus and stops being infectious
when the viral shedding ends.
Data on viral shedding and the infectious period of SARS is very limited. Cori et al. modeled the average
infectious period in SARS patients to be 9.3 days.97 Available literature agrees that viral shedding is low
within the first few days following infection, meaning contagiousness is also low. The available research
from Isakbaeva et al., Cheng et al., and Peiris et al. suggests the viral shedding peaks between day 10 and
day 14 following infection (Supplemental Information on CoV disease course). 98,99,100 However,
Isakbaeva et al. also found viral shedding to persist for 26 days in a patient in the United States.101 The
Centers for Disease Control and Surveillance (CDC) recommends that while SARS patients are most
contagious during their second week of illness, they should also limit contact with others for 10 days after
symptoms subside.102
86 Donnelly CA et al (2003) Epidemiological determinants of spread of causal agent of severe acute respiratory syndrome in
Hong Kong. Lancet 361: 1761-1766
87 Meltzer MI (2004) Multiple contact dates and SARS incubation periods. Emerg Infect Dis 10: 207-209
88 Varia M et al (2003) Investigation of a nosocomial outbreak of severe acute respiratory syndrome (SARS) in Toronto,
Canada. CMAJ 169: 285-292
89 Hsu LY et al (2003) Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial
contacts. Emerg Infect Dis 9: 713-717
90 Leung GM et al (2004) The epidemiology of severe acute respiratory syndrome in the 2003 Hong Kong epidemic: an
analysis of all 1755 patients. Ann Intern Med 141: 662-673
92 ibid.
93 Meltzer MI (2004) Multiple contact dates and SARS incubation periods. Emerg Infect Dis 10: 207-209
94 Varia M et al (2003) Investigation of a nosocomial outbreak of severe acute respiratory syndrome (SARS) in Toronto,
Canada. CMAJ 169: 285-292
95 Hsu LY et al (2003) Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial
contacts. Emerg Infect Dis 9: 713-717
96 Leung GM et al (2004) The epidemiology of severe acute respiratory syndrome in the 2003 Hong Kong epidemic: an
analysis of all 1755 patients. Ann Intern Med 141: 662-673
97 Cori A et al (2009) Temporal variability and social heterogeneity in disease transmission: the case of SARS in Hong Kong.
PLoS computational biology 5: e1000471
98 Isakbaeva ET et al (2004) SARS-associated coronavirus transmission, United States. Emerg Infect Dis 10: 225-231
99 Cheng PK et al (2004) Viral shedding patterns of coronavirus in patients with probable severe acute respiratory syndrome.
Lancet 363: 1699-1700
100 Peiris JS et al (2003) Clinical progression and viral load in a community outbreak of coronavirus-associated SARS
pneumonia: a prospective study. Ibid. 361: 1767-1772
101 Isakbaeva ET et al (2004) SARS-associated coronavirus transmission, United States. Emerg Infect Dis 10: 225-231

4.2.5.3 Symptoms
SARS typically begins with influenza-like symptoms, including high fever, fatigue, sore throat, headache,
and myalgia. Some patients also experience diarrhea, dry cough, and shortness of breath. As SARS
progresses, most cases will develop into pneumonia.103
According to Donnelly et al., the most common symptom is fever, with 94% of cases reporting this
symptom to the Hong Kong Department of Health. Influenza-like symptoms were second most common
at approximately 72% of illnesses. Less than one quarter of patients displayed gastrointestinal symptoms
such as diarrhea, vomiting, and abdominal pain. Approximately 88% of illnesses presented fever plus one
other symptom.104 Without treatment, the pneumonia may lead to respiratory failure and death.
4.2.5.4 Mortality
The overall case fatality-rate of the SARS outbreak is estimated at 11%, according to the WHO. Between
age groups, rates vary from 0%-50%.105 Christian et al. assessed the range to be between 0%- 40% with
an overall fatality rate of 9.6%.106 The high rate of mortality of SARS in the elderly is not accurately
captured by the overall case-fatality rate. Although infection rates were similar, Wang et al. asserts that
the fatality rate in those over 75 years old was 38% whereas no deaths occurred in those under 24
years.107 Similarly, analysis by Donnelly et al. determined the case-fatality rate for persons under 60 years
old to be 6.8% while the rate for over 60 years was 55%.108 Advanced age is the most influential risk
factor for SARS-associated death. In addition to age, diabetes mellitus and hepatitis B virus infection are
other risk factors for death.
4.2.6 MERS-CoV Epidemiology
Middle East Respiratory Syndrome (MERS) is a respiratory infection that can develop into an acute
severe respiratory illness. Many cases end in death, although most who succumb suffer from significant
co-morbidities.
4.2.6.1 Incubation Period
According to the CDC, the incubation period of MERS can range from 2 to 14 days with a median of 5
days. As of July 2015, the WHO supports a median incubation period of 5.5-6.5 days.109 Several
additional literature sources were identified that describe the incubation period. Analysis by Cowling et
al., Assiri et al., and Park et al. determined that the median incubation period of MERS is 6.07 days with a
range from 2 to 15 days (Supplemental Information on CoV disease course). The literature, WHO, and
103 Ibid.
105 World Health Organization. Alert, verification and public health management of SARS in the post-outbreak period.
http://www.who.int/csr/sars/postoutbreak/en/. Last Update August 14, 2003. Accessed July 2015.
106 Christian MD et al (2004) Severe acute respiratory syndrome. Clin Infect Dis 38: 1420-1427
107 Wang MD, Jolly AM (2004) Changing virulence of the SARS virus: the epidemiological evidence. Bull World Health
Organ 82: 547-548
109 The World Health Organization (2015b) Summary of Current Situation, Literature Update and Risk Assessment. Middle
East respiratory syndrome coronavirus (MERS-CoV) 15: 1-7

CDC recommendations concur that most patients begin experiencing symptoms within the first week of
contact with the MERS coronavirus.110,111,112
4.2.6.2 Infectious Period
Limited and inconclusive information is available on the infectious period of MERS. Patients are
considered infectious while they are shedding the virus, but time-specific data is lacking. They are not
contagious during the incubation period, however patients may continue to shed virus after symptoms
have subsided.113,114,115 A study by Memish et al. reported that 76% of studied cases were still shedding
virus 12 days after symptoms appeared. Additionally, analysis showed that sicker patients and those with
significant comorbidities shed MERS-CoV for a longer period of time than standard cases.116
4.2.6.3 Symptoms
MERS symptoms range from mild to severe; patients display symptoms such as fever, cough, sore throat,
shortness of breath, and myalgia that can advance to respiratory failure and septic shock. Approximately
20% of cases have presented as asymptomatic or very mildly symptomatic; it is unknown if asymptomatic
cases are contagious.117
4.2.6.4 Mortality
The case-fatality rate of MERS is estimated at almost 40%.118 There is a clear positive correlation
between increasing age and case-fatality rate. According to the WHO, the median age of MERS cases is
50 years old, with a range from 9 months to 99 years.119 Assiri et al. reported that among cases in Saudi
Arabia, the case-fatality rate was 75% in patients over 60 years of age while there were no fatalities in
patients younger than 19 years.120
Comorbidies also increase a patient’s susceptibility to MERS-CoV. A large percent of MERS fatalities
occur in patients with underlying medical conditions, such as diabetes and hypertension as well as chronic
renal, lung, and cardiac disease.121
110 Cowling BJ et al (2015) Preliminary epidemiologic assessment of MERS-CoV outbreak in South Korea, May–June 2015.
Euro surveillance : bulletin Europeen sur les maladies transmissibles = European communicable disease bulletin 20: 21163
111 Assiri A et al (2013b) Hospital outbreak of Middle East respiratory syndrome coronavirus. N Engl J Med 369: 407-416
112 Park HY et al (2015) Epidemiological investigation of MERS-CoV spread in a single hospital in South Korea, May to June
2015. Euro surveillance : bulletin Europeen sur les maladies transmissibles = European communicable disease bulletin 20:
1-6
113 Cowling BJ et al (2015) Preliminary epidemiologic assessment of MERS-CoV outbreak in South Korea, May–June 2015.
Euro surveillance : bulletin Europeen sur les maladies transmissibles = European communicable disease bulletin 20: 21163
114 The World Health Organization (2014b) Middle East respiratory syndrome coronavirus (MERS‐CoV). WHO Risk
Assesment April 2014: 1-4
115 European Centre for Disease Prevention and Control (2015) Severe respiratory disease associated with Middle East
respiratory syndrome coronavirus (MERS-CoV). Rapid Risk Assessment 20th update: 1-15
116 Memish ZA et al (2014) Middle East respiratory syndrome coronavirus (MERS-CoV) viral shedding in the respiratory tract:
an observational analysis with infection control implications. Int J Infect Dis 29: 307-308
117 The World Health Organization (2015c) Management of asymptomatic persons who are RTPCR positive for Middle East
respiratory syndrome coronavirus (MERS-CoV). Interim guidance July 2015: 1-3
118 Hussain HY (2014) Incidence and Mortality Rate of "Middle East Respiratory Syndrome"-Corona Virus (MERS-Cov),
Threatens and Opportunities. J Mycobac Dis 5.
120 Assiri A et al (2013a) Epidemiological, demographic, and clinical characteristics of 47 cases of Middle East respiratory
syndrome coronavirus disease from Saudi Arabia: a descriptive study. The Lancet Infectious diseases 13: 752-761
121 ibid.

4.3 An overview of GoF Research
This section provides an overview of all Gain-of-Function (GoF) experimental approaches that are
regularly used in the fields of coronavirus and influenza virus research. Our definition of “gain-of-
function” includes all experimental approaches that are reasonably anticipated to lead to one or more of
the following phenotypic changes, as defined in the NSABB’s “Framework for Conducting Risk and
Benefit Assessments of Gain-of-Function Research”:
• Enhanced pathogen production as a result of changes in the replication cycle or growth,
• Enhanced morbidity and mortality in appropriate animal models,
• Enhanced transmission in mammals (e.g., increased host or tissue range, altered route of
transmission, infectivity above a certain threshold determined in an appropriate animal model),
• Evasion of existing natural or induced immunity, and
• Resistance to drugs or evasion of other medical countermeasures such as vaccines, therapeutics,

diagnostics.
These findings are based on two data sources: (1) a comprehensive review of the scientific literature
involving influenza viruses and coronaviruses and (2) interviews with influenza virus and coronavirus
researchers. Within the field of coronavirus research, our literature review included studies involving:
• SARS-CoV,
• MERS-CoV, and
• SARS- or MERS-like animal CoVs, including bat CoVs and civet CoVs.
We did not examine the scientific literature involving the four human coronaviruses that cause mild to
moderate respiratory illnesses such as the common cold or croup: coronaviruses HKU1, OC43, 229E, and
NL63. We note that throughout this report, our use of the term “coronaviruses” or “CoVs” refers
specifically to SARS-CoV, MERS-CoV, and SARS/MERS-like animal CoVs such as HKU4 and HKU5.
We identified approaches involving coronaviruses that are reasonably anticipated to lead to the following
phenotypic changes:

• Altered host range, and
• Evasion of therapeutics in development.
As current animal models for studying coronaviruses do not support transmission between animals, this
field does not include any approaches that lead to enhanced transmission in appropriate animal models.
Additionally, because there is no widespread population immunity to the coronaviruses and there are no
licensed coronavirus vaccines, this field does not include any approaches that lead to evasion of existing
natural or induced immunity. Finally, we did not identify any coronavirus research that is reasonably
anticipated to lead to evasion of diagnostics or of vaccines in development. (We note that there are
currently no FDA-approved vaccines or therapeutics for coronaviruses.)
Within the field of influenza research, our literature review included studies involving:

• Human seasonal strains: currently circulating and historical influenza A H1N1 and H3N2 viruses
and influenza B viruses,
• Human pandemic strains: the 1918 H1N1, 1957 H2N2, 1968 H3N2, and 2009 H1N1 viruses,
• Swine-origin strains: H3N2v and others, and
• Avian-origin strains: H5N1, H7N9, H9N2 and others.
We identified approaches involving influenza viruses that are reasonably anticipated to lead to the
following phenotypic changes:

• Altered host range,
• Enhanced transmission in mammals,
• Evasion of existing natural or induced immunity,
• Evasion of therapeutics, and
• Evasion of vaccines in development.
We note that we are using the term “therapeutics” to include drugs that directly target viruses (e.g.
influenza neuramidase inhibitors), monoclonal antibody-based therapeutics, host immune modulators, and
any other type of anti-viral therapeutic. We did not identify any influenza research that is reasonably
anticipated to lead to evasion of diagnostics.
We note that passaging of influenza viruses and coronaviruses in cells is essential for any experimental
work involving live viruses, both to prepare virus stocks for experimental use and to conduct infection
experiments. This applies to alt-GoF approaches, such as characterization of wild type viruses, as well as
to GoF approaches. Because of the high mutation rates of RNA viruses, including influenza viruses and
coronaviruses, such passaging inevitably selects for higher-yield viruses.122 However, within the
“enhanced virus production” phenotypic category, this analysis is restricted to those approaches that
deliberately seek to enhance virus production through serial passaging, targeted genetic modification, or
other approaches.
Below we briefly summarize the experimental approaches we identified within each phenotypic category,
describing the experimental manipulation, virus strains that are used, and the scientific outcomes of each
approach.
4.3.1 Coronaviruses
4.3.1.1 Enhanced pathogen production as a result of changes in the replication cycle or growth
Serial passaging of CoVs in cell culture leads to the generation of higher-yield viruses. This approach has
been performed using low-yield bat CoV strains to generate higher-yield strains that are suitable for
experimental use. As SARS and MERS naturally grow well in the standard cell culture systems that are
used in the field, researchers are not serially passaging either virus in cell culture to enhance virus
production.
122 Parvin JD et al (1986b) Measurement of the mutation rates of animal viruses: influenza A virus and poliovirus type 1.
Journal of virology 59: 377-383

4.3.1.2 Altered host range
Several experimental approaches alter the host range of CoVs. One approach involves “Spike swapping”
– that is, targeted genetic modification to replace all or part of the coronavirus Spike protein, a viral
surface protein that mediates virus entry into cells and is a critical determinant of host restriction, with the
Spike protein from another CoV species. This manipulation leads to the generation of a recombinant,
chimeric CoV that may exhibit altered host tropism relative to the parental CoV species. The purpose of
these experiments is three-fold:
• Introducing the SARS Spike protein into the backbone of bat CoVs, which do not efficiently
infect standard cell culture lines or animals, enables the chimeric virus to infect cells/animals,
thus creating a tool that can be used to study the biology of the bat CoV,
• Chimeric viruses are used as tools to test whether CoV therapeutics and vaccines are broad-
spectrum, capable of protecting against potentially emerging SARS/MERS-like bat CoVs as well
as SARS and MERS, and
• Testing the ability of chimeric CoVs to infect various types of cells and animals reveals the
breadth of host tropism conferred by a given Spike protein, and comparing the sequences of
parental and donated Spike proteins with different host tropism can uncover amino acid residues
that mediate host restriction.
A second approach involves serial passaging of CoVs in mice, which leads to the generation of viruses
that have adapted to more efficiently infect and cause disease in mice. The purpose of this experiment is
two-fold:
• Mouse-adapted strains are experimental tools that are used for the study of disease pathogenesis
and for testing the efficacy and safety of vaccines and therapeutics, and
• Comparing the sequences of the mouse-adapted and the parental strain leads to the identification
of mutations that are associated with adaptation, which provides a foundation for follow-up
studies investigating the mechanistic basis of virus adaptation to new hosts.
SARS CoV has been passaged in mice by multiple research groups to generate several different mouse-
adapted strains, and chimeric bat-SARS CoVs have also been passaged in mice. Serial passaging of
MERS virus in mice, in order to generate a mouse model for the study of MERS, is ongoing.
A third approach involves targeted mutagenesis to introduce mutations that are associated with altered
host tropism, which has been performed using SARS CoV. These mutations may have been discovered
through a GoF approach, such as serial passaging, or through an alt-GoF approach, such as comparative
sequence analysis. This experiment is performed to demonstrate that the mutation(s) are necessary and
sufficient to alter host tropism, which provides a foundation for follow-up studies investigating the
phenotypic traits underlying virus adaptation to new hosts.
4.3.1.3 Enhanced morbidity and mortality in appropriate animal models
Several experimental approaches enhance the fitness or virulence of CoVs in cell culture or laboratory
animal model systems, respectively. First, serial passaging of CoVs in mice leads to the generation of
viruses with both enhanced infectivity to and virulence in mice. Because of the specificity of virus-host

interactions that are important determinants of host tropism and pathogenicity, this adaptation often
translates to reduced virulence in humans. The purpose of this experiment is two-fold:
• Enhancing the virulence of the virus in mice is an important aspect of creating a mouse model
that replicates human disease pathology, which is needed for the study of disease pathogenesis
mechanisms and the testing of medical countermeasures, and
of mutations that are associated with enhanced virulence, which provides a foundation for follow-
up studies to elucidate the mechanistic basis of virulence. This information can also benefit public
health by identifying new potential targets for therapeutics or for attenuation, in order to create
attenuated vaccine viruses.
A second approach involves targeted genetic modification of viruses to introduce mutations that are
associated with enhanced virulence, which is performed to demonstrate that the mutation(s) are necessary
and sufficient to enhance virulence. As above, this information provides a foundation for follow-up
studies to elucidate the mechanistic basis of virulence.
A final approach involves serial passaging of attenuated viruses in cells or in animals, in order to
determine whether viruses can recover fitness/virulence upon growth in appropriate model systems. This
approach is performed using attenuated viruses that could be used as live attenuated vaccines (LAVs).
Because LAVs with an ability to recover fitness during growth in vivo could cause adverse outcomes in
people, a negative result is an important indicator of safety for any live attenuated vaccine in
development.
4.3.1.4 Evasion of therapeutics in development
Serial passaging of a virus in cells in the presence of a therapeutic may lead to the emergence of viruses
that are resistant to inhibition/neutralization by that therapeutic. This type of experiment has been
performed using SARS CoV, in order to select for escape from monoclonal antibody therapeutics and
other types of therapeutics. The purpose of the experiment is to understand whether and how readily
resistance will arise in response to selective pressure from the therapeutic and to identify mutations that
are associated with resistance to the therapeutic, which provides a foundation for follow-up studies
investigating the mechanisms underlying antiviral activity and antiviral resistance. Because there are no
FDA-approved therapeutics for CoVs, this approach has exclusively been applied to the study of
therapeutics in development.
4.3.2 Influenza viruses
4.3.2.1 Enhanced pathogen production as a result of changes in the replication cycle or growth
Several experimental approaches lead to enhanced production of influenza viruses. The first approach
involves reassortment between a wild type strain and an attenuated, high-yield vaccine backbone strain to
generate a “Candidate Vaccine Virus” (CVV), which comprises the HA and NA genes from the wild type
strain and the remaining six “internal genes” from the vaccine backbone strain and exhibits higher levels
of growth than the parental, wild type virus. CVVs are attenuated and exhibit higher levels of growth
relative to the parental, wild type virus. CVVs may be generated through classical reassortment methods,
which involve co-infection of eggs or cells with the wild type strain and the vaccine backbone strain
followed by antibody-based selection for viruses with the correct surface antigens, or through reverse

genetics.123 These approaches are currently used for the production of influenza vaccines in eggs or cells –
high-yield CVVs serve as the basis for the vaccine strains that are used by manufacturers for large-scale
production. In addition, comparing the sequences of CVVs with different growth properties can lead to
the identification of mutations associated with high growth.
The second approach involves serial passaging of viruses in cells, which selects for higher-yield viruses.
This approach is also a core aspect of the current production of influenza vaccines in eggs or cells.
Specifically, manufacturers serially passage CVVs in eggs or cells to increase CVV yields and to
optimize growth and infection conditions in order to create a vaccine seed strain that is used for large-
scale production of vaccine viruses. The serial passaging approach is also used in academic research,
primarily involving vaccine backbone strains and CVVs but occasionally involving wild type viruses. In
addition to supporting vaccine development, the goals of this experiment could be to identify mutations
associated with high yield, which provides a foundation for follow-up studies investigating the
mechanistic basis of high growth in cells or eggs.
Third, forward genetic screens, which involve random mutagenesis of viruses followed by limited
passaging to select for mutants with high growth properties, enable the identification of mutations that
confer high growth to viruses. Forward genetic screens involving vaccine backbone strains and CVVs
lead to the identification of mutations that are sufficient to enhance the yields of vaccine viruses.
A final approach involves targeted mutagenesis of viruses to introduce mutations that are associated with
high growth. These mutations may have been discovered through a GoF approach, such as serial
passaging or a forward genetic screen, or through an alt-GoF approach, such as comparative analysis of
wild type sequences. This experiment is performed to demonstrate that the mutation(s) are necessary and
sufficient to enhance virus production, which provides a foundation for follow-up studies investigating
the mechanistic basis of the high-growth phenotype.
We note that experimental approaches involving targeted genetic modification of the viral polymerase
complex of avian viruses to render it more “human-like” (through site-directed mutagenesis or
reassortment between human and avian viruses) is also likely to enhance virus replication. However, as
the primary goal of those studies is to gain insight into the mechanisms underlying adaptation of avian
viruses to mammals, we discuss those studies in the “enhanced transmission in mammals” section.
4.3.2.2 Altered host range
Several experimental approaches lead to the generation of viruses with altered host range. First, serial
passaging of viruses in mammalian cells or tissues or in laboratory animals selects for viruses with
enhanced growth in cells or enhanced infectivity to animals, respectively. This type of serial passaging
experiment involves “forced” passaging, meaning that the experimenter directly transfers infected
material, in the form of cell culture supernatant or homogenates of infected tissue, to the subsequent cell
culture dish or animal. Forced serial passaging is carried out for two purposes: (1) to identify mutations
that arise during adaptation of animal-origin viruses (i.e. avian and swine viruses) to mammals, which
provides a foundation for follow-up studies investigating the evolutionary mechanisms driving adaptation
to mammalian hosts and the mechanistic basis of mammalian adaptation, and (2) to develop an mouse
model for the study of a particular virus. Avian and swine viruses are used for studies that seek to
understand the mechanisms underlying mammalian adaptation, and human seasonal viruses are primarily
used for studies that aim to generate new mouse models.
123 Use of classical reassortment methods to generate CVVs may lead to the generation of a 5:3 reassortment strain which
includes the HA, NA, and one additional gene from the wild type strain and the remaining five genes from the vaccine
backbone strain.

A second approach involves deliberate genetic modification of viruses, namely site-directed mutagenesis
and/or reassortment, to introduce genetic traits that may enhance fitness/infectivity in mammals. These
mutations or reassortment gene combinations may be random (i.e. for a forward genetic screen) or may
have been previously shown to be associated with mammalian adaptation or an underlying phenotype,
such as a preference for host receptors decorated with ‘human-like’ sialic acid moieties. Notably, genetic
traits that are associated with mammalian adaptation may be discovered through GoF approaches, such as
serial passaging, or alt-GoF approaches, such as comparative sequence analysis of avian viruses isolated
from human versus poultry infections. Collectively, the deliberate genetic modification approach is used
to discover new genetic traits that contribute to mammalian adaptation and to confirm that particular
genetic traits are necessary and sufficient to enhance fitness/infectivity in mammals. Animal viruses,
including avian and swine viruses, are used exclusively for these studies.
4.3.2.3 Enhanced transmission in mammals
Several experimental approaches lead to the generation of viruses with enhanced transmissibility in
mammals. First, serial passaging of viruses in animals with selection for transmission leads to the
generation of viruses with enhanced transmissibility in mammals. This type of serial passaging
experiment can involve selection for contact transmission, during which the primary (directly inoculated)
and secondary hosts are co-housed, or for airborne transmission, during which the primary and secondary
hosts are separately housed in special isolator cages that prevent direct contact between animals but allow
for air exchange between cages. These studies seek to identify mutations that are sufficient to enhance
transmissibility, which provides a foundation for follow-up studies that investigate the mechanistic basis
of transmissibility in mammals. Animal viruses, including avian and swine viruses, are used exclusively
for these studies.
A second approach involves deliberate genetic modification of viruses, namely site-directed mutagenesis
and/or reassortment, to introduce genetic traits that may enhance transmissibility in mammals. These
mutations or reassortment gene combinations may be random (i.e. for a forward genetic screen) or may
have been previously shown to be associated with transmissibility or an underlying phenotype, such as an
increase in the stability of the HA protein. Notably, genetic traits that are associated with transmissibility
may be discovered through GoF approaches, such as serial passaging, or alt-GoF approaches, such as
forward genetic screens to identify mutations that alter HA stability performed using in vitro, virus-free
systems. Collectively, the deliberate genetic modification approach is used to discover new genetic traits
that contribute to transmissibility and to confirm that particular genetic traits are necessary and sufficient
to enhance transmissibility in mammals. Animal viruses, including avian and swine viruses, are used
exclusively for these studies.
4.3.2.4 Enhanced morbidity and mortality in appropriate animal models
Akin to the enhanced transmission phenotype, both serial passaging and deliberate genetic modification
approaches can lead to the generation of viruses with enhanced morbidity and mortality in appropriate
animal models. Serial passaging of viruses in animals selects for viruses with enhanced virulence and is
used for two purposes: to generate more virulent viruses as a tool for studying the mechanistic basis of
flu-associated morbidity/mortality and for medical countermeasure development, and (2) to identify
mutations that are associated with enhanced virulence, which provides a foundation for follow-up studies
that investigate the mechanistic basis of pathogenicity. These studies can also provide insight into host
mechanisms underlying disease pathology by correlating host immune responses with morbidity and
mortality measures. Both types of serial passaging studies are performed with seasonal and animal (i.e.
avian and swine viruses).

A second approach involves deliberate genetic modification of viruses, through either site-directed
mutagenesis or reassortment, to introduce genetic traits that are expected to enhance pathogenicity. As
above, these mutations or reassortment gene combinations may be random (i.e. for a forward genetic
screen) or may have been previously shown to be associated with a phenotype underlying pathogenicity,
such as evasion of a particular innate immune response. Traits that are associated with enhanced
pathogenicity may be discovered through GoF approaches, such as serial passaging, or alt-GoF
approaches, such as random mutagenesis followed by screening for attenuated virulence (loss-of-
function). Collectively, the deliberate genetic modification approach is used to discover new genetic
traits that contribute to pathogenicity and to confirm that particular genetic traits are necessary and
sufficient to enhance virulence in mammals. These studies are performed using human seasonal viruses,
the 1918 H1N1 pandemic virus, and animal viruses.
We note that the relationship between viral fitness and pathogenicity is complex and that many of the
viral traits that contribute to fitness, either directly or indirectly, mediate pathogenicity. As a result, serial
passaging of viruses in animals may select for both enhanced fitness and enhanced virulence. However,
enhanced viral fitness in vivo does not necessarily translate to high pathogenicity, as seasonal influenza
viruses do not display the morbidity and mortality displayed during infections with zoonotic influenza
viruses such as H5N1, but grow to a high titer.
4.3.2.5 Evasion of existing natural or induced adaptive immunity
Several experimental approaches can lead to the generation of viruses that evade existing natural or
induced immunity. First, serial passaging of viruses in the presence of cognate antibodies may lead to the
acquisition of mutations that allow the virus to escape neutralization by the antibody. This experiment can
be performed in cell culture or in animals that have been vaccinated or previously exposed to influenza
viruses. The second approach involves deliberate modification of the influenza HA protein, the
immunodominant influenza protein and the primary component of influenza vaccine, to introduce
mutations that may lead to antigenic change. In this case, the mutations may be random (i.e. in the context
of a forward genetic screen), previously identified through a GoF approach such as serial passaging, or
previously identified through an alt-GoF approach such as comparative analysis of wild type sequences.
When either approach is performed using recently or currently circulating seasonal influenza viruses or
using seasonal viruses that have recently served as the basis for vaccine strains, the end result is the
generation of a mutant strain that cannot be neutralized by existing natural or induced immunity,
respectively. These studies aim to identify amino acid substitutions that lead to antigenic change and to
define the evolutionary pathways by which those substitutions arise, which provides a foundation for
follow-up studies investigating the evolutionary mechanisms driving antigenic drift and the molecular
basis of antigenic differences between strains.
Because human populations do not have widespread immunity to the 1918 H1N1 pandemic virus or to
animal influenza viruses (i.e. avian viruses and swine viruses), no approaches involving these viruses
meet this phenotypic criterion.
4.3.2.6 Evasion of vaccines in development
Serial passaging of a virus in cells in the presence of animal sera produced in response to a candidate
vaccine or in animals vaccinated with a candidate vaccine may lead to the emergence of viruses that are
resistant to neutralization by that vaccine. This approach is used to test whether and how readily viruses
can evolve to evade vaccines in development, for example new vaccine platforms that are more broad-
spectrum or resistant to drift than current influenza vaccine platforms, which is an important indicator of
the potential field efficacy of the vaccine.

Because seasonal influenza vaccines are updated annually, approaches that lead to the generation of
vaccine strains that are no longer neutralized by vaccine-induced antibodies are more appropriately
described by the “evasion of existing induced immunity” phenotype. In addition, we did not identify any
studies involving H5N1 viruses that would be expected to lead to the generation of viruses that cannot be
neutralized by the pre-pandemic H5N1 vaccine in the national stockpile.
4.3.2.7 Evasion of therapeutics
Several approaches may lead to the generation of viruses that are resistant to therapeutics. The classical
approach involves serial passaging of viruses in the presence of a therapeutic, which may lead to the
acquisition of mutations that allow the virus to evade inhibition by the therapeutic. This approach is
performed to determine whether and how readily a virus evolves resistance in response to selective
pressure from a therapeutic and to identify mutations that confer resistance, which provides a foundation
for follow-up studies investigating the mechanism of action of the therapeutic and the mechanistic basis
of antiviral resistance. Serial passaging approaches have been performed using cell culture, animal
models, and (rarely) human challenge experiments.
The second approach involves deliberate modification of antiviral target proteins to introduce mutations
that may confer antiviral resistance. In this case, the mutations may be random (i.e. in the context of a
forward genetic screen), previously identified through a GoF approach such as serial passaging, or
previously identified through an alt-GoF approach such as comparative analysis of wild type sequences.
Similar to serial passaging experiments, these experiments provide a foundation for follow-up studies
investigating the mechanistic basis of antiviral resistance. Both types of approaches have been performed
using human seasonal viruses, human pandemic strains (i.e. the 1957 H2N2 pandemic virus), and animal-
origin strains. (We note that human challenge experiments have only been performed using human
seasonal strains.)
4.3.2.8 Reassortment
Several experimental approaches can be used to assess the genetic compatibility and fitness of viruses
following reassortment. While the phenotypic consequences of reassortment events between two viruses
cannot be predicted with certainty, reassortant strains may exhibit enhanced fitness, pathogenicity, and/or
transmissibility relative to one or both parental strains. (Notably, reassortant strains may also display
reduced fitness, pathogenicity, and/or transmissibility relative to parental viruses.) In the laboratory,
reassortant viruses can be generated through reverse genetics, which involves the deliberate mixing of
gene segments from two or more viruses in one or multiple combinations, or through co-infection of cells
or animals with two viruses. Follow-up studies may be performed to evaluate pathogenicity, infectivity,
and/or transmissibility of viable reassortants. Collectively, these approaches assess the viability and
phenotypic properties of various reassortment viruses. This information provides a foundation for studies
investigating mechanisms governing reassortment and informs the potential for reassortant viruses to
emerge in nature and the potential public health consequences of such an emergence event.

5 Historical Context of Outbreaks of Influenza, SARS and
MERS
5.1 Purpose and Context 43
5.2 Severe Acute Respiratory Syndrome 43

5.2.1 Summary of findings 43
5.2.2 Background 44
5.2.3 Initiation of the SARS Outbreak 44
5.2.4 Morbidity and Morality 47
5.2.5 Long-term morbidity 49
5.2.6 Economic Burden 50
5.3 Middle East Respiratory Syndrome 51

5.3.1 Summary 51
5.3.2 Background 51
5.3.3 The Emergence of MERS 52
5.3.4 Morbidity and Mortality 53
5.3.5 2015 Outbreak 54
5.4 Influenza 54
5.4.1 Summary 54
5.4.2 Background 55
5.4.3 Seasonal Influenza 55
5.4.4 Morbidity and Mortality of Seasonal Influenza 56
5.4.5 Uncertainty when Determining Influenza-Associated Morbidity and Mortality 56
5.4.6 Consequences in Special Populations 57
5.4.7 Influence of Influenza Virus Type 58
5.4.8 Recent Influenza Seasons 59
5.4.9 Economic Burden of Seasonal Influenza 62
5.4.10 Pandemic Influenza 63
5.4.11 Pandemic Threats 66
5.4.12 Avian Influenza 66
5.4.13 Trends in mortality from influenza in the 20th century 69
5.4.14 Medical Countermeasures Against Influenza 71

5.1 Purpose and Context
In this section, we review the history and current status of influenza, SARS and MERS to provide context
for evaluating the potential risks and benefits associated with GOF studies. Naturally-occurring epidemics
and pandemics present risks to human and animal health and burdens to public health infrastructure. Such
risks are pertinent to the ongoing deliberative process on the risks and benefits of GOF research as they
help to establish the existing risks associated with infectious diseases to which the risks associated with
GOF studies might be compared. This historical perspective will also inform evaluations of the potential
benefit of interventions to reduce the burden of these diseases on society; the greater the harm that these
diseases have inflicted, the greater the potential benefit to society of mitigating their harm.
To the extent available, we have gathered data on past outbreaks of these diseases, and the morbidity,
mortality and economic harm that they have inflicted. For seasonal influenza, these data should provide a
solid baseline for understanding the potential benefits of reducing the burden of this disease that continues
to afflict public health annually. Caution should be used when reviewing the data on outbreaks of
pandemic influenza strains and the human coronaviruses because the disease caused by each new strain
has unique attributes that influence its extent and severity. The next outbreak from a newly emergent
influenza virus or coronavirus could be as severe, not nearly as severe or more severe than any of the
historical outbreaks and there is no science-based means to determine the severity a priori.
For example, although the 1918 influenza outbreak is often held up as the exemplar of the type of
pandemic that researchers are trying to prevent, detect early or mitigate, the severity and extent of the
outbreak may be only partially explained by its unique biology. This pandemic occurred in the waning
years of a world war, when the nutritional status and overall health of the global population was
compromised and when living conditions for the most vulnerable populations were poor (in this case,
younger adults). Perhaps more importantly, public health systems (which rely on the public’s
understanding of the seriousness of infectious disease threats) are far more robust today and therefore
today’s social distancing measures and mass vaccination may greatly mitigate the consequences of an
outbreak. Conversely, the society of the early 20th century was less reliant on complex networks to
provide food, security, water and sanitation services, which could crumble in the face of an outbreak of a
disease that kills a significant number of otherwise healthy adults, leading to secondary deaths from
starvation or social disorder.
5.2 Severe Acute Respiratory Syndrome
5.2.1 Summary of findings
• In 2003, an outbreak of SARS occurred in several Asian countries and Canada, causing nearly
10,000 illnesses and 1,000 deaths, with a disproportionate burden on the elderly,
• Most survivors suffer from long term physical and mental morbidities,
• The outbreak was responsible for $30-100Bn in economic losses, and
• No human cases of SARS have been reported since 2004.

5.2.2 Background
Severe acute respiratory syndrome (SARS) is a viral respiratory disease of zoonotic origin, caused by a
coronavirus identified as SARS-associated coronavirus, or SARS-CoV. Despite ample research, the
natural reservoir of SARS-CoV has not been documented conclusively. The Himalayan masked palm
civet (Paguma larvata), a delicacy in southern China, is commonly attributed as the human transmission
source, but other Chinese delicacies, such as ferret badger (Melogalemoschata), as well as domestic cats
(Felis domesticus), ferrets (Mustela putorius furo), and bats (Rhinolophus) have also been laboratory-
confirmed as virus reservoirs.124
SARS typically begins with flu-like symptoms, including high fever, fatigue, sore throat, headache, and
myalgia. Some patients also experience diarrhea, dry cough, and shortness of breath. As SARS
progresses, most cases will develop into pneumonia. The disease spreads through close contact between
people, mainly by droplet spread of infectious fluids.125 Without treatment, the pneumonia may lead to
respiratory failure and death.
5.2.3 Initiation of the SARS Outbreak
In November 2002, atypical pneumonia of an unknown cause was reported in two patients in Fushan
City, southern Guangdong Province, China. Soon after, similar cases were reported in five other
Guangdong cities.126 The virus remained in China until February 2003 when a physician who had been
treating SARS patients traveled from Guangdong to a hotel in Hong Kong. There he infected ten others
from various countries who then continued traveling, bringing the virus with them to Ireland, Canada,
Vietnam, Singapore, and the United States (Figure 5.1).127
124 The World Health Organization. Severe acute respiratory syndrome. http://www.who.int/topics/sars/en/. Last Update
Accessed July 2015.
Organ 82: 547-548

Figure 5.1. SARS transmission by a patient from Guangdong Province, China to Hong Kong and then to
global travelers reproduced from Christian MD et al.128
These cases sparked a global outbreak. Within weeks, the communicable illness spread to 37 countries
around the world and became recognized as the SARS epidemic of 2003.129 Figure 5.2 below shows the
explosiveness of the outbreak after the international transmission began.130
128 ibid.
Organ 82: 547-548
130 Braden CR et al (2013) Progress in global surveillance and response capacity 10 years after severe acute respiratory
syndrome. Emerg Infect Dis 19: 864-869

Figure 5.2. Reproduced from Chan-Yeung et. al, probable cases of SARS by date of onset or reporting
worldwide.131
SARS is known to transmit through close person-to-person contact and droplet spread, however large,
localized outbreaks suggest additional methods of transmission. Hong Kong’s index case can be traced to
an outbreak in a large apartment complex, Amoy Gardens, where 329 residents became infected.132 The
index case did not contact all three-hundred residents which, with the ability of SARS to remain viable in
feces, presents the possibility of fecal-droplet transmission through the plumbing system of the apartment
complex.133
The CDC also investigated “super spreaders” which are believed to be highly infectious index cases such
as the Guangdong doctor that traveled to a hotel in Hong Kong and the resultant rapid outbreak in
Canada. Possible explanations included a higher SARS-CoV infectious load, aerosolized transmission
that allowed the particles to travel further, and increased age or previous illness that masked the SARS
131 Chan-Yeung M, Xu R-H (2003) SARS: epidemiology. Respirology 8: S9-S14

132 Ibid.

infection. Rapid transmission by super spreaders is credited with initiation and continuation of the SARS
outbreak.134
5.2.4 Morbidity and Morality
According to the World Health Organization (WHO), there were 8,098 SARS cases reported worldwide
during the 2003 epidemic, which resulted in 744 deaths.135 Approximately 30% of cases are believed to
have occurred in healthcare workers due to the necessity of close contact in transmission.136 In the United
States, there were only eight laboratory-confirmed cases and no SARS-associated deaths. All eight
patients had traveled to SARS-infected regions, but did not further transmit the disease upon returning to
the US.137
The WHO estimates the overall case fatality rate of the SARS outbreak to be 11%, with a range of 0%-
50% among age groups.138 Another study estimates the range to be between 0%- 40% with an overall
fatality rate of 9.6%.139
The overall fatality rate, however, minimizes the significant difference that occurred between age groups.
SARS disproportionately kills older people and although the incidence rates did not differ amongst age
groups, the mortality rate in those over 75 years old was 38% whereas no deaths occurred in those under
24 years.140 Advanced age is the most influential risk factor for SARS-associated death. One study
determined the case fatality rate for persons under 60 years old to be 6.8% while the rate for over 60 years
was 55%.141 Another study estimated the average case fatality rate at 45% for persons over 60 years.142
These statistics are provided in Table 5.1 below. All affected regions experienced similar age-specific
trends with a large variance between groups, as can be seen in Figure 5.3 from Anderson et al.
Table 5.1. SARS case fatality rates among age groups

Source Overall < 24 years < 60 years > 60 years > 75 years
Christian et al. 2004143 9.6% - - 45% -
Wang et al. 2004144 - 0% - - 38%
Donnelly et al. 2003145 - - 6.8% 55% -
134 Centers for Disease Control and Prevention. Remembering SARS: A Deadly Puzzle and the Efforts to Solve It. Last Update
April 2013. Accessed
135 Guan Y et al Molecular epidemiology of the novel coronavirus that causes severe acute respiratory syndrome. The Lancet
363: 99-104
136 The World Health Organization (2003b) Severe acute respiratory syndrome (SARS): Status of the outbreak and lessons for
the immediate future. Unmasking a new disease
138 World Health Organization. Alert, verification and public health management of SARS in the post-outbreak period.
http://www.who.int/csr/sars/postoutbreak/en/. Last Update August 14, 2003. Accessed July 2015.
Organ 82: 547-548
143 Ibid.
Organ 82: 547-548

Figure 5.3. Case fatality rates by age and gender in Hong Kong during the 2003 SARS epidemic. Figure
reproduced from Anderson et. al (white bars: male; grey bars: females). 146
Fatality rates also varied between geographical regions. Table 5.2 below, from Chan-Yueng et al.,
displays the number of cases and deaths as well as case fatality rate by country.
Table 5.2. The cumulative number of cases, number of deaths, and case fatality rate by country
during the 2003 SARS epidemic. Table reproduced from Chan-Yeung et.al 147
Country Cumulative number Number of Deaths Case-fatality rate
of cases (%)
Australia 5 0 -
Canada 251 41 17
China 5327 349 7
Hong Kong SAR, China 1755 300 17
Taiwan 346 37 11
Indonesia 2 0 -
Malaysia 5 2 -
New Zealand 1 0 -
Philippines 14 2 -
Korea 3 0 -
Singapore 238 33 14
Thailand 9 2 -
Vietnam 63 2 8
Global 8098 774 9.6
146 Anderson RM et al (2004) Epidemiology, transmission dynamics and control of SARS: the 2002-2003 epidemic.
Philosophical transactions of the Royal Society of London Series B, Biological sciences 359: 1091-1105
147 Chan-Yeung M, Xu R-H (2003) SARS: epidemiology. Respirology 8: S9-S14

Overall mortality rates varied from 0% to 17.1% by region. As the epidemic proceeded, these rates began
to climb. Regions affected early by SARS, such as Guangdong, had case fatality rates ranging from 4% to
10%; regions affected later in the outbreak experienced higher rates, upwards of 13% to 17%.148 One
explanation for this trend is a younger population in regions affected earlier by the virus. Another
possibility is evolution towards a more virulent strain, providing the virus with greater opportunities for
transmission. Worsened symptoms increase the likelihood of droplet spread as well as the need for more
medical treatment and hence, risk of human transmission.149
5.2.5 Long-term morbidity
SARS survivors experienced significant long-term morbidity. Mak et al. used the MOS 36-Item Short
Form (SF-36), a functional outcome assessment, to measure the quality of life of survivors post-illness in
Hong Kong. The study found that SARS survivors performed poorer in all eight quality of life categories
than the normal population (Table 5.3).150 In 2009, the Chinese media reported that 300 survivors
declared continued complications from their illness. Approximately 60% still suffered from medical
issues, including avascular necrosis and pulmonary fibrosis, and 80% could no longer work. Chronic
depression was also reported.151
Table 5.3. Quality of life based on eight domains of the SF-36 assessment between SARS survivors
and the normal population 30 months after the SARS epidemic. Table reproduced from Mak et.
al152
Quality of Life SARS Subjects HK Population Pa
(n=90) normative values
Physical functioning 75.17±22.77 91.83±12.89 <.001**

Role limitations due to
43.54±46.39 82.43±30.97 <.001**
physical health
Bodily pain 58.74±29.98 83.98±21.89 <.001**
General health 40.18±26.58 55.98±20.18 <.001**
Vitality 48.82±22.32 60.27±18.65 <.001**
Social functioning 67.07±27.81 91.19±16.57 <.001**
Role limitations dues to
51.70±46.35 71.66±38.36 <.001**
emotional health
Mental health 61.62±21.57 72.79±16.57 <.001**
Values are mean ± S.D.
a
Two-sided independent sample t test.
** P<.001
Organ 82: 547-548
149 Ibid.
150 Mak IWC et al Long-term psychiatric morbidities among SARS survivors. General Hospital Psychiatry 31: 318-326
151 Xiang YT et al (2014) Outcomes of SARS survivors in China: not only physical and psychiatric co-morbidities. East Asian
archives of psychiatry : official journal of the Hong Kong College of Psychiatrists = Dong Ya jing shen ke xue zhi :
Xianggang jing shen ke yi xue yuan qi kan 24: 37-38

Many SARS survivors suffer from mental morbidity. Several studies have examined the psychiatric status
of survivors at different time points, populations, and locations. Because of these differences, the
percentage of survivors with a psychiatric disorder has fluctuated, however, the burden of mental illness
(Table 5.4). Health care workers also tended to have higher stress levels and depressive symptoms than
non-workers, with 90% qualifying as a potential psychiatric case.153
Table 5.4. Percentage of the measured population with a psychiatric disorder after the SARS epidemic
Source Time After Epidemic Psychiatric Disorder
Lee et al. 2007154 1 year 64%
155
Mak et al. 2009 30 months 33.3%
Lam et al. 2009156 4 years 42.5%
5.2.6 Economic Burden
The SARS epidemic is estimated to have cost anywhere from $30 to $100 billion worldwide from
treatment costs, productivity loss, and decrease in travel and tourism.157 Some estimates place the
economic burden at US $30 billion in the Far East alone.158 This burden translated into a decrease in
Gross Domestic Product (GDP) for many countries. Hong Kong experienced the greatest loss, with a
2.63% decline in GDP, while China’s GDP fell 1.05%. GDP in United States fell 0.07%.159 While the
percentages may appear small, Table 5.5 shows the estimated economic loss in US dollars resulting from
the epidemic.
153 Lee AM et al (2007) Stress and psychological distress among SARS survivors 1 year after the outbreak. Canadian journal
of psychiatry 52: 233
154 Ibid.
156 Xiang YT et al (2014) Outcomes of SARS survivors in China: not only physical and psychiatric co-morbidities. East Asian
archives of psychiatry : official journal of the Hong Kong College of Psychiatrists = Dong Ya jing shen ke xue zhi :
Xianggang jing shen ke yi xue yuan qi kan 24: 37-38
157 Smith RD (2006) Responding to global infectious disease outbreaks: lessons from SARS on the role of risk perception,
communication and management. Soc Sci Med 63: 3113-3123
158 The World Health Organization (2003b) Severe acute respiratory syndrome (SARS): Status of the outbreak and lessons for
the immediate future. Unmasking a new disease
159 (2004) Estimating the Global Economic Cost of SARS. In Learning from SARS: Preparing for the Next Disease Outbreak:
Workshop Summary, Knobler S, Mahmoud A, Lemon S, Mack A, Sivitz L, Oberholtzer K (eds). Washington (DC)

Table 5.5. Estimates of the economic consequences of the SARS epidemic by region and worldwide (in
USD)
Source Region Economic Loss
Mackenzie et al. 2013160 Worldwide $40 billion
161
Lee et al. 2004 Worldwide $40- $54 billion
Wen et al. 2004162 China $25.3 billion
163
Siu et al. 2004 China $48 billion
Fan 2003164 East & Southeast Asia $12.3- $28.4 billion
On May 18, 2004, the WHO declared the last SARS case to be contained. No human infections have been
reported since due to the stringent disease control and public health response.165
5.3 Middle East Respiratory Syndrome
5.3.1 Summary
• Outbreaks of MERS were first identified in 2012 and have occurred sporadically since that time,
• To date these outbreaks have led to more than 1,000 cases and nearly 500 deaths, the burden of
which fell disproportionately on the elderly with significant co-morbidities, and
• The vast majority of cases have been identified in the Middle East, and, recently South Korea
although cases have been identified sporadically in countries in Europe and North America.
5.3.2 Background
Middle East Respiratory Syndrome (MERS) is an acute severe respiratory infection of zoonotic origin
caused by a corona virus knows as MERS-CoV. Although the natural source is unconfirmed, MERS-CoV
has been identified in camels in the Arabian Peninsula. MERS transmission has occurred through close
contact between humans and animals, most likely dromedary camels or bats.166 Both the MERS-CoV
virus and its antibodies have been isolated in dromedary camels in the Arabian Peninsula. These findings
reinforce the possibility of camel-to-human transmission through close contact and the ingestion of raw
camel milk. While camel meat could also be a source of infection, cooking the meat is customary and
inactivates the virus.167
160 Mackenzie JS, Merianos A (2013) The legacies of SARS - international preparedness and readiness to respond to future
threats in the Western Pacific Region. Western Pacific surveillance and response journal : WPSAR 4: 4-8
161 (2004) Estimating the Global Economic Cost of SARS. In Learning from SARS: Preparing for the Next Disease Outbreak:
Workshop Summary, Knobler S, Mahmoud A, Lemon S, Mack A, Sivitz L, Oberholtzer K (eds). Washington (DC)
162 Wen H et al (2004) The Short-Term Impact of SARS on the Chinese Economy. Asian Economic Papers 3: 57-61
163 Siu A, Wong YCR ibid.Economic Impact of SARS: The Case of Hong Kong. 62-83
164 Fan, Emma Xiaoqin. 2003. SARS: Economic Impacts and Implications. © Asian Development Bank.
http://hdl.handle.net/11540/616. License: CC BY 3.0 IGO.
165 Mackenzie JS, Merianos A (2013) The legacies of SARS - international preparedness and readiness to respond to future
threats in the Western Pacific Region. Western Pacific surveillance and response journal : WPSAR 4: 4-8
167 Abdel-Moneim AS (2014) Middle East respiratory syndrome coronavirus (MERS-CoV): evidence and speculations. Arch
Virol 159: 1575-1584

Patients display symptoms such as fever, cough, shortness of breath, sore throat, and myalgia; some
infected people show no symptoms at all. Approximately 3 to 4 out of every 10 persons suspected of
MERS have died. Most fatalities, however, had significant co-morbidities that exacerbated the MERS
symptoms.168
5.3.3 The Emergence of MERS
The first MERS case was reported in Saudi Arabia in September 2012. From September 2012 to July
2015, 1,368 laboratory-confirmed cases have been reported to the World Health Organization (WHO), at
least 487 of which resulted in death. A total of 26 countries in the Middle East, Africa, Europe, Asia, and
North America have had at least one MERS case. Approximately 75% of cases, however, have been from
Saudi Arabia. All cases have been directly or indirectly linked to the Middle East. The United States has
only had two cases, both in travelers. 169 This information can be seen in Figure 5.4 below from the WHO.
Figure 5.4. Confirmed MERS cases around the world since 2012. Figure reproduced from The World Health
Organization, Summary of Current Situation.170
168 The Centers for Disease Control and Prevention. Middle East Respiratory Syndrome (MERS).
http://www.cdc.gov/coronavirus/mers/. Last Update June 22, 2015. Accessed July 2015.
170 The World Health Organization. Middle East respiratory syndrome coronavirus (MERS-CoV) maps and epicurves.
http://www.who.int/csr/disease/coronavirus_infections/maps-epicurves/en/. Last Update July 2015. Accessed July 2015.

5.3.4 Morbidity and Mortality
The overall MERS case fatality rate is almost 40%.171 The WHO reported the median age of SARS cases
to be 50 years old, with a range from 9 months to 99 years.172 One study of MERS patients in Saudi
Arabia found a clear relation between case fatality rates and increasing age. The data supporting this
correlation is provided in Table 5.6 below.173
Table 5.6. Case fatality rates with increasing age174

Age Case fatality rate
< 19 years 0%
< 50 years 39%
< 60 years 48%
> 60 years 75%
Studies have also found that a large proportion of MERS cases are in patients with underlying medical
conditions, such as diabetes and hypertension as well as chronic renal, lung, and cardiac disease (Table
5.7).175
Table 5.7. Percent of MERS patients with underlying medical conditions

Source Percent with Comorbidity
Penttinen et al. 2013176 73%
177
Assiri et al. 2013 96%
Arabi et al. 2014178 100%
Although the disease may have existed for some time before detection, MERS is a relatively new
communicable disease. Research on long-term morbidity, mortality, and economic burden of the illness is
in progress as the epidemic itself is still ongoing. Cases have been localized mostly to the Middle East,
although cases occurred in China, Thailand, the Philippines, and the Republic of Korea in the spring and
summer months of 2015.
171 Hussain HY (2014) Incidence and Mortality Rate of "Middle East Respiratory Syndrome"-Corona Virus (MERS-Cov),
Threatens and Opportunities. J Mycobac Dis 5.
174 Ibid.
175 Ibid.
178 Arabi YM et al (2014) Clinical course and outcomes of critically ill patients with Middle East respiratory syndrome
coronavirus infection. Ann Intern Med 160: 389-397

5.3.5 2015 Outbreak
As of the writing of this section (summer of 2015), the Republic of Korea is currently experiencing the
largest known MERS outbreak outside of the Arabian Peninsula. In May 2015, an infected traveler
returned from the Middle East with MERS. His travel history went unreported for over a week, openly
exposing many people to the virus. 179 Since then, there have been 186 confirmed cases of MERS—185 in
the Republic of Korea and one in China—of which the median age is 55 years old.180 There have been 36
deaths reported; 91.7% of the deaths were in the elderly or patients with co-morbidities. Approximately
17,000 people were quarantined.181
The Republic of Korea’s cultural traditions are said to have influenced the rapid transmission of MERS
within the country. Customs encourage friends and family to not only visit ill patients, but provide
extensive bedside care. Further, in Korea, patients tend to visit several medical facilities before admitting
themselves. These dispositions increase exposure and transmission of the virus.182 It is not surprising
therefore that all of the cases in Korea have been associated with health care facilities; 14% were in
medical professionals. Every patient has been connected to the index case and no cases have occurred in
the general population.183 The last MERS case was reported to the WHO on July 4, 2015 and officials
believe the outbreak to under control in China and the Republic of Korea. Cases continue to be reported,
however, in the Arabian Peninsula.184
5.4 Influenza
5.4.1 Summary
• Influenza viruses cause human outbreaks seasonally, in pandemics and via zoonotic infection
from birds,
• Morbidity and mortality from influenza is difficult to measure because death is due to secondary
causes,
• Five to ten percent of the population worldwide gets influenza every year, and it is associated
with 250,000-500,000 deaths annually, the burden of which falls mostly on the elderly,
• Seasonal influenza causes approximately $100Bn in direct and indirect economic losses in the US
annually,
• Pandemic strains have caused a handful of outbreaks in the last 100 years, and are sometimes
associated with significantly more illness or deaths than seasonal strains,
179 The World Health Organization. Middle East respiratory syndrome coronavirus (MERS-CoV).
http://www.who.int/emergencies/mers-cov/en/. Last Update 2015. Accessed July 2015.
180 World Health Organization. Middle East respiratory syndrome coronavirus (MERS-CoV) – Republic of Korea.
http://www.who.int/csr/don/17-july-2015-mers-korea/en/. Last Update July 17, 2015. Accessed July 2015.
181 World Health Organization (2015) Briefing for Foreign Correspondents MERS Outbreak.
182
Ibid..
183 World Health Organization. Middle East respiratory syndrome coronavirus (MERS-CoV) – Republic of Korea.
http://www.who.int/csr/don/17-july-2015-mers-korea/en/. Last Update July 17, 2015. Accessed July 2015.
184 The World Health Organization. Middle East respiratory syndrome coronavirus (MERS-CoV) maps and epicurves.
http://www.who.int/csr/disease/coronavirus_infections/maps-epicurves/en/. Last Update July 2015. Accessed July 2015.

• Pandemic strains have been noted for disproportionally harming groups other than the elderly
(young adults for the 1918 pandemic and young children for the 1957 and 2009 pandemics),
• Deaths from pandemics and seasonal influenza decreased across the 20th century until the 1960s,
after which time the death rate from influenza has remained relatively constant,
• Outbreaks of avian influenza have caused up to $20B in direct and indirect economic losses due
to destruction of poultry flocks and lost trade,
• Avian influenza outbreaks have an unpredictable effect on human health, the worst are associated
with up to 1,000 cases and 500 deaths, some cause mild illness in humans, and
• Vaccines and antivirals demonstrate significant efficacy in clinical trials, however their overall
public health benefit is more difficult to measure.
5.4.2 Background
Influenza is a highly contagious viral infection of the respiratory tract caused by three orthomyxoviruses
of different antigenic types —influenza A, B, and C. These influenza viruses can cause seasonal and
pandemic outbreaks. According to the CDC, an influenza pandemic “can occur when a non-human
(novel) influenza virus gains the ability for efficient and sustained human-to-human transmission and then
spreads globally.” Seasonal outbreaks, however, occur annually on predictable seasonal patterns and are
caused by recirculating influenza viruses with residual immunity among the population. Both seasonal
and pandemic outbreaks spread through human-to-human transmission. Sporadically, avian influenza
viruses, which do not normally infect or transmit through humans, will cross species barriers and cause an
outbreak in the human population.185
Influenza type A can infect a variety of animal hosts and is further divided into subgroups based on its
surface proteins (e.g. H1N1, H3N2, H5N1). This genetic variation allows type A viruses to cause
pandemic outbreaks, dominate seasonal epidemics, and cross species barriers.186 Type B viruses have a
more limited host range and limited variation, and therefore, do not cause pandemic outbreaks. Virus type
C causes only mild symptoms in humans and does not contribute to epidemics.187 Influenza virus is
consistently one of the leading causes of illness in the United States and is associated with significant
mortality; it is among the US Centers for Disease Control and Prevention’s (CDC) top priorities.188
5.4.3 Seasonal Influenza
Seasonal influenza viruses circulate through the population causing annual epidemics during the winter
months in temperate climates, such as the United States, and unpredictable epidemics in tropical regions.
The World Health Organization (WHO) estimates that 5%-10% of adults and 20%-30% of children
185 Centers for Disease Control and Prevention. Influenza (Flu). http://www.cdc.gov/flu/. Last Update 2015. Accessed May
2015.
186 Centers for Disease Control and Prevention. Types of Influenza Viruses. http://www.cdc.gov/flu/about/viruses/types.htm.
Last Update 2014. Accessed May 2015.
187 Ibid.
188 Centers for Disease Control and Prevention. Influenza (Flu). http://www.cdc.gov/flu/. Last Update 2015. Accessed May
2015.

worldwide are infected with influenza each year. Of those illnesses, 3 to 5 million develop into severe
cases, which result in 250,000 to 500,000 deaths annually.189
5.4.4 Morbidity and Mortality of Seasonal Influenza
In developed countries, approximately 1.2 in 10,000 persons die annually as a result of influenza.190
Industrialized countries with established surveillance systems provide the majority of case data so these
attack rates could be substantially underestimated for developing, tropical countries with limited financial
and technical resources.191 Globally, influenza drives the loss of 19.2 million disability-adjusted life years
(DALYs) annually (16.9 million-21.5 million), as estimated in the Global Burden of Disease Study in
2010. This statistic is equivalent to 279 DALYs (245-311 DALYs) per 100,000 people worldwide.192
According to the National Strategy for Pandemic Influenza by the Homeland Security Council, an
average of over 200,000 hospitalizations and 36,000 deaths are caused each year by seasonal influenza in
the US alone.193 Anywhere from 5-20% of the population, or 15 to 60 million Americans, is infected with
the influenza virus annually.194 Death tolls range from 1.4 to 16.7 deaths per 100,000 persons, exceeding a
total of 49,000 lives.195 A CDC study estimated that, since the 1980s, the hospitalization rate has ranged
from 150,000 to 431,000 people per year.196 Seasonal influenza is consistently ranked in the top ten
overall leading causes of death in the United States.197 After convalescence, however, no long term
morbidities are associated with influenza.198
5.4.5 Uncertainty when Determining Influenza-Associated Morbidity and Mortality
The CDC is uncertain on exactly how many people become infected with or die from influenza each year.
States are required to report influenza-related deaths in children under 18 years old only, leaving all other
cases and deaths untracked. Seasonal influenza is rarely listed as a cause of death on adults’ death
certificates. Additionally, death often occurs weeks after initial infection when patients are no longer
symptomatic and the virus cannot be detected. Some diagnostic tests can even produce false negative
results due to their low sensitivity.199 For these reasons, the CDC must estimate the number of annual
cases and deaths caused by influenza. The estimate is typically based on deaths related to pneumonia and
influenza (P&I) to account for the fact that although influenza is never confirmed as the cause of death,
pneumonia is most often the underlying cause.200 Excess P&I deaths, however, only account for
approximately 25% of actual influenza-related mortality.
189 World Health Organization. Influenza (Seasonal). http://www.who.int/mediacentre/factsheets/fs211/en/. Last Update March
2014. Accessed May 2015.
190 Lagace-Wiens PR et al (2010) Influenza epidemiology--past, present, and future. Crit Care Med 38: 1-9
191 World Health Organization (2005) Influenza Vaccines. WHO Position Paper 33: 279-287
192 Institute for Health Metrics and Evaluation (2013) The Global Burden of Disease: Generating Evidence, Guiding Policy. 1-
50
193 Gerberding JL, Centers for Disease Control and Prevention. (2005) Avain Influenza: Preparing for a possible Influenza
Pandemic
194 Fiore AE et al (2010) Prevention and control of influenza with vaccines : recommendations of the Advisory Committee on
Immunization Practices (ACIP), 2010, Atlanta, Ga . Dept. of Health and Human Services, Centers for Disease Control and
Prevention.
195 Centers for Disease C, Prevention (2010) Estimates of deaths associated with seasonal influenza --- United States, 1976-
2007. MMWR Morbidity and mortality weekly report 59: 1057-1062
196 Thompson WW et al (2004) Influenza-associated hospitalizations in the United States. Jama 292: 1333-1340
197 Centers for Disease Control and Prevention. (2015c) Leading Causes of Death.
198 Rothberg MB, Haessler SD (2010) Complications of seasonal and pandemic influenza. Crit Care Med 38: e91-97
199 Centers for Disease Control and Prevention. Estimating Seasonal Influenza-Associated Deaths in the United States: CDC
Study Confirms Variability of Flu. http://www.cdc.gov/flu/about/disease/us_flu-related_deaths.htm. Last Update March
2015. Accessed May 2015.

One estimation method for influenza deaths is excess mortality, which is the difference between mortality
rates during an influenza epidemic and the standard baseline rate in the absence of an influenza outbreak.
While there is never a total absence of influenza illness, comparing the rate of illness during a predictable
seasonal outbreak to the baseline rate during the summer months when there is no outbreak provides an
estimation for the illnesses being caused by the outbreak. For example, during the 2013-2014 influenza
season the percentage of deaths attributed to P&I peaked at 8.7%. The baseline rate of P&I related deaths
that occur during a non-outbreak time period, however, was about 7% and hence, the difference of 1.7%
of P&I related deaths can be attributed to the seasonal influenza outbreak.201
The excess index can also be applied to the number of outpatient visits to evaluate excess morbidity.202
The national baseline for the percentage of outpatient visits is the mean percentage of P&I visits during
non-outbreak weeks for the previous three seasons plus two standard deviation and is usually about 2.0%.
The baseline for mortality is calculated using a periodic regression model applied to data from the
previous five years. It ranges from 6-8%, but is usually around 7.5% during seasonal influenza months.203
Excess morbidity and mortality are not intended to provide exact statistics, but serve as an indicator of the
season’s relative severity. Hospitalization rates are not reported as an excess index, but instead as
laboratory-confirmed statistics.
5.4.6 Consequences in Special Populations
Influenza harms young children and the elderly more than older children and non-elderly adults. The
WHO reports that excess mortality attributed to influenza ranges from three to 15 per 10,000 Americans
older than 65 years. In the general population, excess morbidity is approximately 1.2 in 10,000 persons
per year.204 Excess hospitalization averages at 100 per 10,000 for children under six months old, but only
four per 10,000 children once they are more than five years old.205 Persons in the age group from 5 to 49
years have the lowest hospitalization rate.206
Seasonal influenza disproportionally kills the elderly, who suffer approximately 80-90% of the mortality
observed.207 This trend is evident in Figure 5.5, from a CDC sponsored study on the epidemiology of
seasonal influenza, giving both hospitalization and mortality rates.
Influenza cases typically resolve within two weeks, however severe cases or cases in elderly or at-risk
populations may lead to additional complications. Complications include pneumonia, bronchitis, and
exacerbation of existing pulmonary and respiratory diseases and could lead to death. Resolved cases,
however, are not associated with long-term morbidity.208
201 Centers for Disease Control and Prevention. (2014b) Influenza Activity — United States, 2013–14 Season and Composition
of the 2014–15 Influenza Vaccines. Morbidity and Mortality Weekly Report, Vol. 63, pp. 483-490.
202 Simonsen L (1999) The global impact of influenza on morbidity and mortality. Vaccine 17 Suppl 1: 3-10
203 Centers for Disease Control and Prevention. Overview of Influenza Surveillance in the United States. Last Update Accessed
June 2015.
207 Dawood FS et al (2012) Estimated global mortality associated with the first 12 months of 2009 pandemic influenza A H1N1
virus circulation: a modelling study. Lancet Infectious Diseases 12: 687-695
208 Rothberg MB, Haessler SD (2010) Complications of seasonal and pandemic influenza. Crit Care Med 38: e91-97

Figure 5.5. Rates of hospitalization and mortality across age groups of seasonal influenza from 1976-2000 in
the US Figure reproduced from Thompson et. al.209
Table 5.8 and Table 5.9 below present several studies’ estimated annual mortality rate and hospitalization
rate, respectively, of seasonal influenza.
Table 5.8. Studies on the influenza-associated mortality rate (per 100,000 persons) across age
groups of seasonal influenza annually in the US
Source All ages < 1 year < 5 years < 65 years ≥ 65 years
Thompson et al. 2003210 3.1 0.3 0.2 - 22.1
211
Simonsen et al. 2000 2.5 - - 0.49 18.7
CDC 2010212 2.4 - - 0.4 17.0
Table 5.9. Studies on the influenza-associated hospitalization rate (per 100,000 persons) across age
groups of seasonal influenza annually in the US
Source All ages < 1 year > 5 years < 65 years ≥ 65 years
213
Thompson et al. 2004 37 - 26.3 13 205
Simonsen et al. 2000214 49 - - 33 174
215
Zhou et al. 2012 - 151 94 - 309
5.4.7 Influence of Influenza Virus Type
Since 1968, influenza A, both H1N1 and H3N2, and influenza B have all co-circulated to cause seasonal
epidemics. Seasons dominated by influenza A H3N2 tend to cause greater excess mortality than influenza
209 Thompson WW et al (2006) Epidemiology of seasonal influenza: use of surveillance data and statistical models to estimate
the burden of disease. J Infect Dis 194 Suppl 2: S82-91
210 Thompson WW et al (2003) Mortality associated with influenza and respiratory syncytial virus in the United States. Jama
289: 179-186
211 Simonsen L et al (2000) The impact of influenza epidemics on hospitalizations. J Infect Dis 181: 831-837
212 Centers for Disease C, Prevention (2010) Estimates of deaths associated with seasonal influenza --- United States, 1976-
2007. MMWR Morbidity and mortality weekly report 59: 1057-1062
214 Simonsen L et al (2000) The impact of influenza epidemics on hospitalizations. J Infect Dis 181: 831-837
215 Zhou H et al (2012) Hospitalizations associated with influenza and respiratory syncytial virus in the United States, 1993-
2008. Clin Infect Dis 54: 1427-1436

A H1N1 and B seasons.216 From 1990-1999, overall influenza-related mortality ranged from 17,000 to
51,000 deaths in a season. Around 90% of seasons were predominated by H3N2 strains and thus,
influenza-associated mortality rose during that time compared to an average year. The increase, however,
can also be partly associated with the large proportion of the population that was over 65 years old and
tend to suffer more from influenza infections.217
5.4.8 Recent Influenza Seasons
In the past decade, both seasonal and pandemic strains have caused fluctuating outbreaks, as can be seen
in Figure 5.6 and Figure 5.7 below. From 2007-2008, the H3N2 seasonal virus predominated in the
United States. That season saw the greatest mortality and hospitalization rates of the previous four years,
at 83 deaths of children under four years old. The percentage of the population with P&I related deaths
peaked at 9.1%, which is well above the normal baseline of 7.6%.218
The following influenza season, from 2008-2009, was less severe and can be attributed to the H1N1
dominant strain that typically causes more mild outbreaks. There were only 45 pediatric deaths and a
hospitalization rate of 2.8 per 10,000 children. Outpatient visits peaked at 3.7%, well over the baseline of
2%, and the weekly percentage of deaths attributed to P&I peaked at 7.6%, which is at the baseline of
7.6%.219
A pandemic virus emerged the following year. The 2009 H1N1 pandemic strain was unusually severe,
causing an extended outbreak from April 2009 through May 2010. Infections persisted through the
summer months, speaking to the strain’s augmented pathogenicity and transmissibility. Excess mortality
peaked at 8.1% in November and again at 8.2% in January, exceeding the national baseline for thirteen
weeks straight. The pandemic resulted in the greatest number of patient visits of any year since influenza
surveillance began in 1997.220 Seasonal influenza typically burdens the elderly population the most with
80-90% of the mortality. This pandemic, however, affected children and young adults much more
substantially. An estimated 80% of deaths were in people under 65 years old.221 There were 344 reported
pediatric deaths.222
By the 2010-2011 season, the pandemic H1N1 strain continued to circulate. The H3N2 strain and
influenza B were also widely distributed, resulting in a more balanced attack rate among all age groups.
The season was significantly less severe than the previous pandemic year, but still worse than the
previous 2008-2009 endemic. There were 105 confirmed pediatric deaths. Outpatient visits peaked at
217 Fiore AE et al. (2007) Prevention and Control of Influenza : Recommendations of the Advisory Committee on
Immunization Practices (ACIP). Morbidity and Mortality Weekly Reports Recommendations and Reports. Centers for
Disease Control and Prevention, Atlanta, GA, Vol. 56, pp. 1-54.
218 Centers for Disease Control and Prevention. (2008) Influenza Activity --- United States and Worldwide, 2007--08 Season.
Morbidity and Mortality Weekly Report, Vol. 57, pp. 692-697.
219 Centers for Disease Control and Prevention. (2009) Update: Influenza Activity-- United States, September 28, 2008- April
4, 2009, and Composition of the 2009-2010 Influenza Vaccine Morbidity and Mortality Weekly Report, Vol. 58, pp. 369-
374.
220 Centers for Disease Control and Prevention. (2010) Update: Influenza Activity United States, 2009--10 Season. Morbidity
and Mortality Weekly Report, Vol. 59, pp. 901-908.

4.6%, the hospitalization rate was 4.38 per 10,000 children, and excess mortality surpassed the epidemic
threshold for 13 consecutive weeks, peaking at 8.9%.223
Once again in the 2011-2012 season, all three strains—H1N1, H3N2, and type B— circulated, but H3N2
was the predominant virus. This mild season was noted for lower hospitalization rates, outpatient visits,
and deaths than previous years. The hospitalization rate amounted to only 0.86 per 10,000 people and
patient visits dropped to the lowest since point 1997 by meeting, but not exceeding the threshold baseline
of 2.4%. The P&I associated death rate surpassed the baseline for one short week, at 7.9%, and only 26
pediatric deaths were reported.224
The 2012-2013 epidemic was more severe. All three virus types circulated, but H3N2 was the main
circulating virus. Patient visits soared past endemic thresholds for 15 weeks at a high of 6.1%, much
greater than the national baseline of 2%, and hospitalization rates escalated to 4.43 per 10,000 people.
Mortality also surpassed the baseline for 13 weeks, reaching 9.9%. There were 149 influenza-associated
pediatric deaths.225
From 2013-2014, the pandemic influenza A H1N1 strain co-circulated with limited H3N3 and B strains.
This was the first season since 2009 that the pandemic strain predominantly recirculated, this time as a
seasonal virus. Although there were fewer deaths and hospitalizations than typically observed with an
H1N1 season, adults were once again at a higher-risk for influenza. Those aged 50-64 experienced 5.43
hospitalizations per 10,000 persons compared to the aggregate rate of 3.56 per 10,000 across all age
groups. Outpatient visits exceeded the baseline for 15 consecutive weeks, peaking at 4.6%, and excess
mortality exceeded for 8 weeks, maxing out at 8.7%. There were 96 pediatric deaths reported.226
Table 5.10 below compiles key CDC statistics from the Emerging Infections Program (EIP) for recent
influenza seasons in the United States. Beginning in 2009, EIP was expanded to surveille 26 million more
Americans, 8.5% of the population.227 This new FluSurv-NET program may account for some of the
differences in comparable statistics.
223 Centers for Disease Control and Prevention. (2011) Update: Influenza Activity --- United States, 2010--11 Season, and
Composition of the 2011--12 Influenza Vaccine. Morbidity and Mortality Weekly Report, Vol. 60, pp. 705-712.
224 Centers for Disease Control and Prevention. (2012) Update: Influenza Activity — United States, 2011–12 Season and
Composition of the 2012–13 Influenza Vaccine. Morbidity and Mortality Weekly Report, Vol. 61, pp. 414-420.
of the 2013–14 Influenza Vaccine. Morbidity and Mortality Weekly Report, Vol. 62, pp. 473-479.

Table 5.10 Influenza-associated statistics as reported to the CDC annually for past influenza seasons
in the United States. Hospitalization rate is given per 100,000 persons
Season Dominant Outpatient P&I Hospitalization Hospitalization Pediatric
Strain Visits Related Rate ≤ 4 years Rate ≥ 65 Deaths
Mortality years
2007-2008228 H3N2 6.0% 9.1% 40.3 - 83
2008-2009229 H1N1 3.7% 7.6% 28.0 10.0 45
2009-2010230 pH1N1 7.6% 8.2% 83.0 32.0 344
H3N2/
2010-2011231 4.6% 8.9% 43.8 62.5 105
H1N1
2011-2012232 H3N2 2.4% 7.9% 14.2 30.4 26
2012-2013233 H3N1 6.1% 9.9% 66.2 191.2 149
234
2013-2014 H1N1 4.6% 8.7% 46.9 88.1 96
Figure 5.6 below from the CDC presents the percentage of influenza-associated outpatient visits in the US
by surveillance week over several past influenza seasons. Figure 5.7 depicts P&I attributable deaths in
influenza seasons since 2009. Both figures plot statistics in regards to their national baselines so that
excess morbidity and mortality can be visualized. The fluctuation between yearly seasons and pandemic
outbreaks are apparent.
228 Centers for Disease Control and Prevention. (2008) Influenza Activity --- United States and Worldwide, 2007--08 Season.
Morbidity and Mortality Weekly Report, Vol. 57, pp. 692-697.
229 Centers for Disease Control and Prevention. (2009) Update: Influenza Activity-- United States, September 28, 2008- April
4, 2009, and Composition of the 2009-2010 Influenza Vaccine Morbidity and Mortality Weekly Report, Vol. 58, pp. 369-
374.
232 Centers for Disease Control and Prevention. (2012) Update: Influenza Activity — United States, 2011–12 Season and
Composition of the 2012–13 Influenza Vaccine. Morbidity and Mortality Weekly Report, Vol. 61, pp. 414-420.
of the 2013–14 Influenza Vaccine. Morbidity and Mortality Weekly Report, Vol. 62, pp. 473-479.

Figure 5.6. Percentage of outpatient visits attributable to influenza by surveillance week during past influenza
seasons in the United States, as reported to the CDC. Figure reproduced from Centers from Disease Control
and Prevention, Influenza Activity — United States, 2013–14 Season and Composition of the 2014–15
Influenza Vaccines.235
Figure 5.7. Percentage of P&I attributable deaths by surveillance week during influenza seasons since 2009 in
the United States. Figure reproduced from Centers from Disease Control and Prevention, Influenza Activity
— United States, 2013–14 Season and Composition of the 2014–15 Influenza Vaccines.236
5.4.9 Economic Burden of Seasonal Influenza
The US public health infrastructure is burdened with 24.7 million (19.9-30.1 million) cases of seasonal
influenza annually. These cases generate approximately 31.4 million (22.6-43.5 million) outpatient visits
235 Ibid.
236 Ibid.

and 3.1 million days of hospitalization, producing approximately a $10.4 billion ($4.1- $22.2 billion)
burden of direct medical costs from annual influenza endemics, according to a 2003 study by the
Immunization Service Division and Division of Viral Diseases at the CDC.237 Additionally, anywhere
from $8.7 to $31 billion in earnings is lost annually due to decreases in productivity and loss of life from
the estimated 610,000 (360,000- 953,000) DALYs lost. In total, the annual costs of seasonal influenza
amount to $87.1 billion ($47.2- $149.5 billion) in direct and indirect costs in the United States.238 Table
5.11 below displays several studies’ estimated economic harm of seasonal influenza on the US. Echoing
these data, the WHO reported that France and Germany may spend anywhere from $1 million to $6
million per 100,000 people annually on influenza outbreaks.239
Table 5.11. Studies on the direct medical cost and total economic burden of seasonal influenza annually on
the US
Source Medical Costs (per year) Economic Burden (per year)
240
Office of Technology Assessment 1981 $1- $3 billion -
Molinari et al. 2007241 $10.4 billion $87.1 billion
242
Mao et al. 2012 $10.3 billion $29.1 billion
5.4.10 Pandemic Influenza
In addition to the annual burden of seasonal influenza, several pandemic strains have caused additional
morbidity and mortality over the past century. Over the past 300 years, ten pandemics are known to have
occurred, three of which were in the 20th century.243
The 1918 influenza pandemic, also known as the “Spanish Flu” (the exact location of origin was never
determined), was the most deadly outbreak in modern history.244 The H1N1 outbreak occurred in three
waves beginning in March 1918. The second and most severe wave occurred concurrently in North
America, Africa, and Europe in August 1918.245 Over the next six months, anywhere from 20-40% of the
global population was infected with influenza and approximately 500 million people became ill.246
Estimates on the total mortality ranges from 20 to 100 million worldwide, but most estimates suggest
237 Molinari NA et al (2007) The annual impact of seasonal influenza in the US: measuring disease burden and costs. Vaccine
25: 5086-5096s
238 Ibid.
240 US Congress: Office of Technology Assessment: Cost-effectiveness of Influenza Vaccination. Washington, DC: GPO;
1981.
241 Molinari NA et al (2007) The annual impact of seasonal influenza in the US: measuring disease burden and costs. Vaccine
25: 5086-5096
242 Mao L et al (2012) Annual economic impacts of seasonal influenza on US counties: spatial heterogeneity and patterns. Int J
Health Geogr 11: 16
243 Osterholm MT (2005) Preparing for the next pandemic. N Engl J Med 352: 1839-1842
245 Ibid.

approximately 50 million deaths.247, 248 In the United States, between 550,000 and 675,000 deaths were
attributed to the pandemic.249, 250
Unlike seasonal influenza, the highest rates of morbidity and mortality were observed in young, healthy
adults instead of the elderly.251 The case fatality rate was greater than 2.5%, compared to later pandemics
that were less than 0.1%.252 The reason for the severity of the outbreak is unclear, considering the attack
rate and age distribution was similar to other pandemics. Moreover, H1N1 outbreaks are typically
associated with lower levels of morbidity and mortality.253 In total, anywhere from 1% to 3% of the global
population died as a result of this pandemic.254
In February 1957 another pandemic strain emerged, influenza A H2N2 originating from China. The virus
spread rapidly around the world and to the United States, with most related excess mortality occurring
between September 1957 and March 1958.255 Morbidity in children exceeded 50%.256 Approximately
70,000 deaths occurred in the United States and two million worldwide, which amounts to approximately
0.07% of the population worldwide dying from influenza-associated causes.257
Just over a decade later, a new influenza A H3N2 strain emerged in Hong Kong in July 1968. The virus
was slow moving; it didn’t reach the United States until December and Europe the following year.258
Attack rates peaked at 40% in children, but mortality rates were highest in the elderly. The excess
mortality amounted to approximately 33,800 deaths in the United States, mild for a pandemic outbreak.259
The 1968-1969 outbreak had the lowest mortality rate of any other pandemic of the century, possibly due
to partial immunity from exposure to the 1957 pandemic strain and improved medical treatment.260
Overall, approximately 0.03% of the population worldwide died from influenza-associated causes during
the 1968 pandemic.261
Table 5.12 below presents the age distribution of mortality attributable to influenza during 20th century
influenza pandemics. Figure 5.8 shows antigenic type and associated age-based mortality of pandemics
through 1995. Interpandemic seasons are also included for comparison.
248 Noymer A, Garenne M (2000) The 1918 influenza epidemic's effects on sex differentials in mortality in the United States.
Population and development review 26: 565-581
250 Crosby, A. 1989. America's Forgotten Pandemic: The Influenza of 1918. Cambridge: Cambridge University Press.
254 Murray CJ et al (2006) Estimation of potential global pandemic influenza mortality on the basis of vital registry data from
the 1918-20 pandemic: a quantitative analysis. Lancet 368: 2211-2218
256 Cox NJ, Subbarao K (2000) Global epidemiology of influenza: past and present. Annual review of medicine 51: 407-421
257 Mathews JD et al (2009) Understanding influenza transmission, immunity and pandemic threats. Influenza and other
respiratory viruses 3: 143-149
259 Cox NJ, Subbarao K (2000) Global epidemiology of influenza: past and present. Annual review of medicine 51: 407-421
261 Mathews JD et al (2009) Understanding influenza transmission, immunity and pandemic threats. Influenza and other
respiratory viruses 3: 143-149

Table 5.12. Influenza-associated mortality (per 100,000) across age groups during
pandemics in the US 262
Season Pandemic Strain All ages < 65 years ≥ 65 years
1918-1919 H1N1 529 546 166
1957-1958 H2N2 39 15 273
1968-1969 H3N2 8.1 4.3 44
Figure 5.8. Age distribution of deaths associated with influenza A pandemics and interpandemic seasons in
the United States, 1918-1995, reproduced from Simonsen et. al263
In 2009, a pandemic H1N1 strain quickly circulated through 74 countries during the summer months, an
unusual time for the virus.264 The excess mortality is estimated between 152,000 and 575,000 people
worldwide. Populations in Africa and Asia suffered approximately 50% of the deaths, possibly due to the
limited availability of medical treatment, presence of underlying health conditions, lower qualities of care,
and fragile nutritional status.265 In the initial stages of the pandemic, Mexico was spending an estimated
$57 million per day trying to control and treat the virus. The World Bank claimed a total $3 trillion loss
from the burden of the pandemic.266
In the United States, from 43 to 89 million people became infected with pandemic H1N1, resulting in
9,000 to 18,300 deaths. Point of care testing used to identify the virus is less sensitive on pandemic strains
262 Simonsen L et al (1998) Pandemic versus epidemic influenza mortality: a pattern of changing age distribution. J Infect Dis
178: 53-60
263 Ibid..

and therefore, infection rates also could have been much greater.267 Children experienced significantly
higher mortality and hospitalization rates than the elderly, who are normally the most affected population
group, as can be seen in Table 5.13. The hospitalization rate was 8.3 per 10,000 among ages 0-4 years and
3.4 per 10,000 among ages 5-17 years while only 3.2 per 10,000 for those over 65 years.268 The severity
and extent of the outbreak was curtailed by the rapid vaccination of more than 80 million people. The
H1N1 pandemic strain continues to circulate today as a seasonal human flu virus.
Table 5.13. Statistics on the number of influenza cases, influenza-related mortality, and
hospitalization rate resulting from the 2009 H1N1 pandemic in the US
Hospitalization Rate (per 10,000 persons)
Season Case Count Death Toll
< 4 years 5-17 years > 65 years
2009 43-89 million 9,000-18,000 8.3 3.4 3.2
5.4.11 Pandemic Threats
Over the past century, there were also several instances of newly emerged influenza strains that were
feared to cause a pandemic but did not. In 1976, a swine strain with similarities to the 1918 pandemic
strain emerged at Fort Dix, New Jersey. Due to a robust vaccination campaign and other unknown factors,
the virus never became widespread. Then again in May of 1977, a new virus type surfaced in China and
began to rapidly spread around the world. The strain was similar to strains circulating prior to 1957. Many
adults had already developed immunity to the virus, which limited outbreaks to children mainly and
prevented a major pandemic. Decades later in 1997, once again in China, a novel H5N1 virus began to
infect young adults directly from chickens. Over one million chickens were culled to successfully prevent
further spread of the avian flu.269
5.4.12 Avian Influenza
Within the past two decades, avian influenza A has caused substantial morbidity and mortality in humans.
Of the many existing strains, H5 and H7 subtypes have been the primary transmission sources from birds
to humans. H5N1 viruses are endemic in Asia and Africa while subtype H7 circulates across Europe and
North America.270 Avian influenza A viruses are either highly pathogenic (HPAI) or have low
pathogenicity (LPAI) based on the degree of infection caused by their molecular characteristics. While
bird-to-human transmission of avian influenza is a persistent issue, sustained human-to-human
transmission of avian strains has not been observed.271
The H5N1 virus first infected humans in Hong Kong in 1997 stemming from an outbreak in poultry. It
reemerged in mainland China in 2003 and quickly spread through wild bird migration and domestic
poultry trading among Asia, Africa, and Europe. The virus remains endemic in birds, with sporadic
270 Belser JA et al (2009) Past, Present, and Possible Future Human Infection with Influenza Virus A Subtype H7. Emerging
271 Peiris JS et al (2007) Avian influenza virus (H5N1): a threat to human health. Clin Microbiol Rev 20: 243-267

transmission to humans.272 These H5N1 viruses had various genotypes of geographically-related
sublineages. 273 Since reemerging in 2003, and up to July 2015, there have been 844 cases of human
infection with H5N1 and 449 deaths reported to the WHO, equating to a 53% mortality rate (Figure 5.9).
Sixteen countries have had outbreaks of H5N1 avian influenza; Egypt, Indonesia, and Viet Nam have
experienced the most cases.274
H5N1 was the first avian influenza virus to continuously circulate in Asia for over 16 years.275 According
to the Food and Agriculture Organization of the United Nations, by 2008 the economic losses were
estimated to have already reached $20 billion worldwide. The two US outbreaks were estimated to have
cost $65 and $140 million from the loss of poultry and cost of disease control alone. The United States
also committed $1.4 billion towards the international effort against H5N1.276 The economies of countries
in East and Southeast Asia suffered the most due to the affect H5N1 had on the poultry industry. From
2003 to 2005, an estimated 10 billion dollars was lost from the death or culling of over 140 million birds
in Southeast Asia. GDP was depleted by 0.6% to 2% among affected countries.277
Figure 5.9. Number of confirmed H5N1 cases in humans by month and country as of July 2015 reproduced
from The Centers for Disease Control and Prevention, Summary and assessment as of 17 July 2015.278
In March 2013, the H7N9 virus was identified as the causative agent of infections in two patients in
Shanghai and one in Anhui Province, China. Spread throughout China continued through exposure to
272 The World Health Organization. Avian Influenza Fact Sheet.

http://www.who.int/mediacentre/factsheets/avian_influenza/en/. Last Update March 2014. Accessed August 2015.
274 The World Health Organization. Cumulative number of confirmed human cases for avian influenza A(H5N1)
http://www.who.int/influenza/human_animal_interface/EN_GIP_20150717cumulativeNumberH5N1cases.pdf?ua=1. Last
Update July 2015. Accessed August 2015.
275 Simms L, Jeggo M (2014) Avian influenza from an ecohealth perspective. EcoHealth 11: 4-14
276 Commission of the European Communities (2015) Impact Assessment Avian Influenza. 171
278 The World Health Organization (2015e) Summary and assessment as of 17 July 2015. Influenza at the human-animal
interface: 1-4

infected birds, typically in live poultry markets, but not through human-to-human contact.279 As of July
2015, 677 cases of H7N9 with at least 275 deaths have been reported to the WHO, a 41% mortality rate
(Figure 5.10). The virus remains endemic in China, but no cases have been reported outside of the
mainland.280 According to the United Nations, China experienced over $6.5 billion in losses from the
H7N9 outbreak.281
Figure 5.10. Number of confirmed H7N9 cases in humans by week in China as of July 2015. Figure
reproduced from The World Health Organization, Summary and assessment as of 17 July 2015. 282
Since 2002, other influenza H7 subtypes have caused more than 100 human infections. Subtype H7N2
caused a widespread outbreak on domestic turkey farms in the northeastern United States in 2002. One
worker in Virginia became infected with the virus while birds were being culled to contain the outbreak,
which proved the possibility of bird to human transmissibility in the US. A year later in New York, H7N2
was identified in an immunocompromised man who denied contact with any live poultry.283 The
transmission source remains unknown.284 Both US cases of human H7N2 fully recovered. In the United
Kingdom, several cases of H7N2 were reported in 2007; three people were hospitalized, all of whom
recovered.285
279 The World Health Organization. Avian Influenza Fact Sheet.

http://www.who.int/mediacentre/factsheets/avian_influenza/en/. Last Update March 2014. Accessed August 2015.
interface: 1-4
281 Nebehay, S. (2013) China's bird flu outbreak cost $6.5 billion. Reuters.
interface: 1-4
283 Belser JA et al (2009) Past, Present, and Possible Future Human Infection with Influenza Virus A Subtype H7. Emerging
284 Belinda Ostrowsky et al., “Low Pathogenic Avian Influenza A (H7N2) Virus Infection in Immunocompromised Adult, New
York, USA, 2003,” Emerging Infectious Diseases 18, no. 7 (July 2012): 1128–31
285 E. M. Abdelwhab, J. Veits, and T. C. Mettenleiter, “Prevalence and Control of H7 Avian Influenza Viruses in Birds and
Humans,” Epidemiology and Infection 142, no. 5 (May 2014): 896–920

Human H7N3 has been identified in Italian poultry workers in 2003, Canadian poultry workers in 2004,
and one poultry worker in the United Kingdom in 2006. 286,287,288 Years later in 2012, H7N3 caused a
severe outbreak in chicken farms in Mexico that resulted in two infected workers. While H7N3 is capable
of human infection, few cases and no deaths have been reported; the subtype’s capability within humans
appears limited.289
The first emergence of H7N7 was in 1996 when a woman became ill after cleaning her poultry shed in
England.290 Then in 2002, a H7N7 outbreak in the Netherlands became the first avian influenza outbreak
in humans since H5N1 emerged. A poultry outbreak on commercial farms led to more than 1000 people
with subclinical indications, 86 human infections, and at least one death.291 During an outbreak in Italy in
2013, three poultry workers contracted H7N7 without respiratory symptoms. Human to human
transmission did not occur.292 Transmissibility of the H7N7 strain is still not well understood.
5.4.13 Trends in mortality from influenza in the 20th century
Unlike the newly emergent diseases SARS and MERS, influenza has caused human misery for centuries.
Because of the lack of historical data as well as the difficulty of determining influenza mortality as
previously described, little research exists on the historical trends of influenza. As discussed above (and
shown in Figure 5.12), although Simonsen, et al focused on pandemic influenza, their data on mortality
from intrapandemic seasons shows an overall downward trend in mortality from seasonal influenza.293
Doshi et. al used mortality reports from Vital Statistics of the United States along with US Census
estimates to calculate the incidence of influenza mortality over the past century during pandemic and
seasonal outbreaks (Figure 5.11).294 The Health Sentinel also analyzed mortality reports from Vital
Statistics of the United States along with other statistics from the HHS to determine the incidence of
influenza mortality over the past century (Figure 5.12).295 Although both groups likely capture deaths
from respiratory ailments other than influenza, both groups show a significant reduction in annual deaths
from influenza. Importantly, although both papers suggest an overall drop in mortality over the past
hundred years, both papers also clearly show that the downward trend has largely ceased, as mortality
from influenza has been roughly the same in the last thirty years (and perhaps over the last 50). Without
significant medical advances in the future, we can expect seasonal influenza to kill tens of thousands of
Americans and nearly half a million people globally every year.
286 Puzelli et al., “Serological Analysis of Serum Samples from Humans Exposed to Avian H7 Influenza Viruses in Italy
between 1999 and 2003,” The Journal of Infectious Diseases 192, no. 8 (October 15, 2005): 1318–22
287 Tweed et al., “Human Illness from Avian Influenza H7N3, British Columbia,” Emerging Infectious Diseases 10, no. 12
(December 2004): 2196–99
288 Nguyen-Van-Tam et al., “Outbreak of Low Pathogenicity H7N3 Avian Influenza in UK, Including Associated Case of
Human Conjunctivitis,” Euro Surveillance: Bulletin Européen Sur Les Maladies Transmissibles = European Communicable
Disease Bulletin 11, no. 5 (2006): E060504.2.
289 Lopez-Martinez I et al (2013a) Highly pathogenic avian influenza A(H7N3) virus in poultry workers, Mexico, 2012. Emerg
Infect Dis 19: 1531-1534
290 Kurtz, et. al, “Avian Influenza Virus Isolated from a Woman with Conjunctivitis,” Lancet (London, England) 348, no. 9031
(September 28, 1996): 901–2
291 Enserink, “Infectious Diseases. Bird Flu Infected 1000, Dutch Researchers Say,” Science (New York, N.Y.) 306, no. 5696
(October 22, 2004): 590
292 Puzelli S et al (2014a) Human Infection with Highly Pathogenic A(H7N7) Avian Influenza Virus, Italy, 2013. Emerging
293 Simonsen L et al (1998) Pandemic versus epidemic influenza mortality: a pattern of changing age distribution. J Infect Dis
178: 53-60
294 Puzelli S et al (2014a) Human Infection with Highly Pathogenic A(H7N7) Avian Influenza Virus, Italy, 2013. Emerging
Infectious Diseases 20: 1745-1749-945
295 Health Sentinel. The World Health Organization. Severe acute respiratory syndrome. http://www.who.int/topics/sars/en/.
Last Update Accessed July 2015.

Figure 5.11. Crude influenza-classed mortality per 100,000 persons by month from 1900–2004 (top) and 1930–2004 (bottom) in the United States, as
reproduced from Doshi et. al.296
296 Puzelli S et al (2014a) Human Infection with Highly Pathogenic A(H7N7) Avian Influenza Virus, Italy, 2013. Emerging Infectious Diseases 20: 1745-1749-945

Figure 5.12. United States mortality rate for influenza and pneumonia from 1900-2002, reproduced from The
Health Sentinel.297
5.4.14 Medical Countermeasures Against Influenza
Medical countermeasures (MCM) are crucial for both preparedness and response to influenza. There are
two main types of influenza countermeasures: antiviral agents and vaccinations. Research on
development of an influenza vaccine began soon after the virus was isolated in 1933.298 The first wide-
scale use of the vaccine occurred in 1945 during World War II among the US military. In 1960, after the
1957-1958 pandemic, the US Surgeon General recommended influenza vaccinations for high risk groups,
including the elderly, pregnant women, and those with chronic conditions. Then in 2010, after the 2009
H1N1 pandemic, the Advisory Committee on Immunization Practices promoted universal influenza
vaccination in persons over 6 months for the first time.299
Adamantanes and neuraminidase inhibitors are the two classes of antiviral drugs approved by the Food
and Drug Administration (FDA) for use against influenza. The adamantanes, amantadine and
rimantadine, are agents against influenza A and were approved for treatment in the 1966 and 1973,
respectively. The Centers for Disease Control and Prevention (CDC) no longer recommends adamantanes
for the treatment of seasonal influenza due to increasing resistance of circulating strains.300 The
297 Health Sentinel. The World Health Organization. Severe acute respiratory syndrome. http://www.who.int/topics/sars/en/.
Last Update Accessed July 2015.
298 Hannoun C (2013) The evolving history of influenza viruses and influenza vaccines. Expert review of vaccines 12: 1085-
1094
299 Osterholm MT et al (2012) Efficacy and effectiveness of influenza vaccines: a systematic review and meta-analysis. The
Lancet Infectious diseases 12: 36-44
300 Centers for Disease Control and Prevention. Use of Antivirals: Background and Guidance on the Use of Influenza Antiviral
Agents. http://www.cdc.gov/flu/professionals/antivirals/antiviral-use-influenza.htm. Last Update Feb 25, 2015. Accessed
October 2015.

neuraminidase inhibitors, oseltamivir and zanamivir were both approved in 1999 and function against
influenza A and B. 301 Peramivir, also a neuraminidase inhibitor, was recently approved in December 2015.302
5.4.14.1 Efficacy
Seasonal influenza and pandemic influenza respond differently to the available MCM. This section
summarizes the efficacy of the currently available MCM. Additional details and references are provided
in the supplemental information.
Antivirals
There are two categories of influenza antiviral drugs, but because adamantanes are no longer
recommended for treatment of seasonal influenza, they are not considered in this analysis.303
There are three neuraminidase inhibitors used to treat seasonal influenza: oseltamivir, zanamivir, and
peramivir. Antiviral treatment is most effective when administered within 48 hours of symptom onset.304
Antivirals can have a variety of benefits for patients suffering from seasonal influenza (see Supplemental
Information on Antiviral and Vaccine Efficacy):
• Duration of symptoms is decreased by 15-40%,

• Reduction of the probability of death from influenza by five-fold, and
• Reduction in the amount of viral titer and the duration of viral shedding by 25-70%.
Antivirals can be provided as prophylaxis as well. If provided prior to exposure, antivirals prevent
symptoms of seasonal influenza in 80% of patients.
Vaccines
Based on the CDC’s reported adjusted overall vaccine effectiveness (the reduction in risk of needing a
doctor’s visit) for influenza seasons from 2005-2015 (excluding the 2008-2009 influenza season), the
weighted average of vaccine effectiveness for seasonal influenza is 44.2%.305 Seasonal influenza
vaccination is also effective in preventing severe influenza. Each year, on average, vaccination produces,
a 15% reduction in hospitalizations due to influenza illness. This result agrees with more recent studies
that also show less than a 50% efficacy.306
One study suggests that a previous year’s influenza vaccine may confer protection against current
circulating viruses. During the 2012-2013 season, the vaccine effectiveness against influenza A (H3N2)
301 Department of Health and Human Services NIoH. (2006) Development and Use of Antivirals for Pandemic Influenza
Meeting Summary Bethesda, MD
302 U.S. Food and Drug Administration. (2014) FDA approves Rapivab to treat flu infection. FDA News Release. U.S. Food and
Drug Administration.
303 Centers for Disease Control and Prevention. Use of Antivirals: Background and Guidance on the Use of Influenza Antiviral
Agents. http://www.cdc.gov/flu/professionals/antivirals/antiviral-use-influenza.htm. Last Update Feb 25, 2015. Accessed
304 Ibid.
305 Centers for Disease Control and Prevention. Seasonal Influenza Vaccine Effectiveness, 2005-2015.
http://www.cdc.gov/flu/professionals/vaccination/effectiveness-studies.htm. Last Update June 24, 2015. Accessed Aug 11,
2015.
306 Cowling, BJ et al “Assessment of influenza vaccine effectiveness in a sentinel surveillance network 2010-13, United
States.” Vaccine 2015, article in press.

among people who received the 2012-2013 vaccination was similar to those who received only the 2011-
2012 vaccination.307
5.4.14.3 Pandemic H1N1 Influenza
Antivirals
During the outbreak of pandemic H1N1 influenza, most patients were treated with oseltamivir, based on
the CDC’s recommendation. Of 3,362 2009 pandemic influenza A (H1N1) virus isolates collected, only
three did not show resistance to adamantanes.308 Patients that received oseltamivir treatment had a
survival rate of 90.3%, and antiviral treatment was associated with a 20% reduced mortality risk when
compared to no treatment.309 In addition, patients receiving early oseltamivir treatment had shorter fever
durations than patients who did not receive antiviral treatment.310,311
Antiviral treatment is also effective in reducing viral titer and duration of viral shedding in pH1N1
patients. According to one study, antiviral treatment reduced the mean viral load of patients by 14.3%.312
On average, antiviral treatment reduced the duration of viral shedding in patients by 34% when compared
to untreated patients, or patients who received antiviral treatment after 48 hours, which has been shown to
be less effective.313
A study in ferrets show that prophylaxis with oseltamivir did not prevent H1N1 infection.314 However, a
household contact study showed that contacts under 20 years old who received antiviral prophylaxis with
oseltamivir or zanamivir were nearly seven-fold less likely to be infected with pandemic H1N1 influenza
than those who did not receive antiviral prophylaxis (odds ratio 0.15).315 Additionally, other observational
studies indicate that prophylaxis may be effective in preventing H1N1 infection.316,317
307 McLean HQ et al (2015) Influenza vaccine effectiveness in the United States during 2012-2013: variable protection by age
and virus type. The Journal of infectious diseases 211: 1529-1540
308 Gubareva LV et al (2010) Comprehensive assessment of 2009 pandemic influenza A (H1N1) virus drug susceptibility in
vitro. Antiviral therapy 15: 1151-1159
309 Muthuri SG et al (2014) Effectiveness of neuraminidase inhibitors in reducing mortality in patients admitted to hospital with
influenza A H1N1pdm09 virus infection: a meta-analysis of individual participant data. The Lancet Respiratory medicine 2:
395-404
310 Yu H et al (2010) Effectiveness of oseltamivir on disease progression and viral RNA shedding in patients with mild
pandemic 2009 influenza A H1N1: opportunistic retrospective study of medical charts in China. BMJ 341: c4779
311 Li IW et al (2010) The natural viral load profile of patients with pandemic 2009 influenza A(H1N1) and the effect of
oseltamivir treatment. Chest 137: 759-768
312 Meschi S et al (2011) Duration of viral shedding in hospitalized patients infected with pandemic H1N1. BMC infectious
diseases 11: 140
313 Nicholson KG et al (2000) Efficacy and safety of oseltamivir in treatment of acute influenza: a randomised controlled trial.
Neuraminidase Inhibitor Flu Treatment Investigator Group. Lancet 355: 1845-1850
314 Oh DY et al (2014) Evaluation of oseltamivir prophylaxis regimens for reducing influenza virus infection, transmission and
disease severity in a ferret model of household contact. The Journal of antimicrobial chemotherapy 69: 2458-2469
315 Odaira F et al (2009) Assessment of secondary attack rate and effectiveness of antiviral prophylaxis among household
contacts in an influenza A(H1N1)v outbreak in Kobe, Japan, May-June 2009. Euro surveillance : bulletin Europeen sur les
maladies transmissibles = European communicable disease bulletin 14
316 Asiedu-Bekoe F et al (2012) Mass oseltamivir prophylaxis halts pandemic influenza A H1N1 2009 outbreak in a secondary
school in Ashanti Region, Ghana. Ghana medical journal 46: 219-224
317 Leung YH et al (2011) A school outbreak of pandemic (H1N1) 2009 infection: assessment of secondary household
transmission and the protective role of oseltamivir. Epidemiology and infection 139: 41-44

Vaccines
Monovalent pH1N1 influenza vaccine demonstrated an average effectiveness of 66% (range 60-
93%).318,319
5.4.14.4 Epidemiological Evidence of Effectiveness
The previous section describes the efficacy of vaccination in controlled, clinical trials. The
epidemiological evidence is equivocal.
Supporting the benefit of vaccination, the Influenza Division at the CDC released a study on the mortality
of the nine influenza seasons from 2005 to 2014. Their analysis found that approximately 22% of
influenza-associated deaths were prevented by the influenza vaccine, with 90% of this benefit realized
persons over 65 years. 320 Further, they conservatively estimate 40,000 deaths to have been averted. The
study found that the fewest deaths were prevented during the 2009-2010 pandemic.321 Another CDC study
found that from 2005 to 2011, influenza-associated illnesses and hospitalizations were substantially
alleviated by vaccinations. The analysis estimated that 1.1 to 5 million illnesses and 7,700 to 40,400
hospitalizations were prevented in those vaccinated against influenza. The study also found the largest
benefit to be in elderly populations.322
Other studies found less favorable outcomes. Demicheli et al. presented a meta-analysis that showed poor
effectiveness of the vaccine in reducing influenza cases and work days lost in healthy adults. The study
did find however, significant reductions in serologically confirmed cases of influenza.323 According to the
National Center for Health Statistics, while vaccination coverage spanned 15-20% of the elderly
population by 1980 and increased to 65% in 2001, influenza-associated mortality, however, substantially
increased among those over 65 years old.324 Likewise, a retrospective analysis covering the years 1979 to
2000 from The Institute for Chronic Illnesses found the vaccine to have little or no effectiveness for
preventing influenza cases, deaths, or hospital admissions.325 Simonsen et al. at the National Institute of
Allergy and Infectious Diseases published that based on national vital statistics, other studies on the
effectiveness of the influenza vaccine overestimate its benefits.326 Despite concurrent increasing
vaccination coverage, Rizzo et al. found no evidence of reduction in influenza related morality from 1970
to 2001 in the Italian elderly population.327 Several research groups propose that the overestimate of
vaccine benefits is due to unrecognized confounding variables, such as a disproportionate amount of
318 Griffin MR et al (2011) Effectiveness of non-adjuvanted pandemic influenza A vaccines for preventing pandemic influenza
acute respiratory illness visits in 4 U.S. communities. PLoS One 6: e23085
320 Foppa IM et al (2015) Deaths averted by influenza vaccination in the U.S. during the seasons 2005/06 through 2013/14.
Vaccine 33: 3003-3009
321 ibid.
322 Kostova D et al (2013) Influenza Illness and Hospitalizations Averted by Influenza Vaccination in the United States, 2005-
2011. PLoS One 8: e66312
323 Demicheli V et al (2004) Vaccines for preventing influenza in healthy adults. Cochrane Database Syst Rev: CD001269
324 Simonsen L et al (2005) Impact of influenza vaccination on seasonal mortality in the US elderly population. Arch Intern
Med 165: 265-272
325 Geier et al (2006) Influenza Vaccine: Review of Effectiveness of the U.S. Immunization Program, and Policy
Considerations. Journal of American Physicians and Surgeons
326 Simonsen L et al (2005) Impact of influenza vaccination on seasonal mortality in the US elderly population. Arch Intern
Med 165: 265-272
327 Rizzo C et al (2006) Influenza-related mortality in the Italian elderly: no decline associated with increasing vaccination
coverage. Vaccine 24: 6468-6475

healthy elderly persons being vaccinated.328 Moreover, Madea et al., did not find significantly different
levels of infection between vaccinated and unvaccinated healthy children 6 to 24 months of age.329
Regardless of differing opinions on vaccine efficacy, the decline in influenza morbidity and mortality
over the past century is irrefutable. As Doshi et al. highlights, however, influenza vaccination was not
available until the 1940s and not widely adopted until the 1980s and hence cannot be responsible for the
drop (see Figures 5.11 and 5.12 above).330
Few studies exist on the relationship between influenza countermeasures and the decreases observed in
morbidity and mortality, however the advent of antibiotics likely significantly reduced risk of death from
influenza. In the absence of antibiotics, the majority of influenza mortality is attributed to interactions
between the influenza virus and bacteria colonizing the upper respiratory tract, causing fatal secondary
infections.331 Evidence strongly suggests that the Great Pandemic would have been greatly mitigated with
the availability of antibiotics.332
5.4.14.5 Availability of Vaccines
Vaccine shortages have captured headlines several times over the past couple decades for various reasons,
including underestimation of demand, reduction in manufacturers, contaminated issues, and unexpected
outbreaks.333 After the severe influenza vaccine shortages of the 2004 to 2005 season, the United States
Government Accountability Office completed a study on the status of seasonal influenza preparation. The
final report recognized that the vaccine shortage of 4.7 million doses, approximately half of the needed
supply, exposed the need for better preparation for seasonal endemics.334 During the 2009 H1N1
pandemic, experts predicted 160 million doses of the pandemic vaccine would be available for public
vaccination by October, yet only 30 million were delivered by that date.335
329 Maeda T et al (2004) Failure of inactivated influenza A vaccine to protect healthy children aged 6-24 months. Pediatrics
international : official journal of the Japan Pediatric Society 46: 122-125
330 Doshi P (2008) Trends in recorded influenza mortality: United States, 1900-2004. Am J Public Health 98: 939-945
331 Morens DM et al (2008) Predominant role of bacterial pneumonia as a cause of death in pandemic influenza: implications
for pandemic influenza preparedness. J Infect Dis 198: 962-970
332 Doshi P (2008) Trends in recorded influenza mortality: United States, 1900-2004. Am J Public Health 98: 939-945
333 Tosh PK et al (2010) Influenza Vaccines: From Surveillance Through Production to Protection. Mayo Clinic Proceedings
85: 257-273
334 United States Government Accountability Office. Shortages in 2004-2005 season underscore need for better preparation
(2005) Rep. No. GAL-05-984: 1-5
335 Vaccines Ho. Influenza Pandemics. http://www.historyofvaccines.org/content/articles/influenza-pandemics. Last Update
July 2014. Accessed October 2015.

6 Risk Assessment of Laboratory Accidents and Natural
Disasters
6.1.1 Seasonal Influenza Viruses 79
6.1.2 Pandemic Influenza Viruses 81
6.1.3 Avian Influenza Viruses 82
6.2 Methodology 85
6.2.1 Purpose of This Task 85
6.2.2 Input from Modeling Subject Matter Experts 86
6.2.3 Interplay of the Components of the Biosafety Risk Assessment 86
6.2.4 Taking a Parametric Approach to the Analysis 88
6.2.5 Probability of an Infection Outside of Containment 90
6.2.6 Predicting the probability that an outbreak escapes local control 94
6.2.7 Modeling the Global Consequences of a Human-Transmissible Outbreak 95
6.2.8 Simplified Modeling of Bird-Transmissible Pathogens 96
6.2.9 Estimating Risk of Experiments Involving GoF Pathogens 98
6.3 Practices in GoF Laboratories that Reduce Risk but Are Not Included in Our Study 100
6.3.1 Training 100
6.3.2 Exercises and Drills 101
6.3.3 Laboratory Practices 101
6.3.4 Health Precautions 103
6.3.5 Institutional Culture 104
6.3.6 Additional Institutional Policies 104
6.4 Probability of Laboratory Acquired Infections 105

6.4.1 The selection of incidents to include in the RBA 105
6.4.2 Identification of locations at risk of earthquakes and floods 108
6.4.3 Irreducible uncertainty prevents an accurate prediction of absolute risk 108
6.4.4 Relative Probability of Laboratory Acquired Infections 109
6.5 Consequences of an Outbreak Caused by an Avian Influenza Strain that Is Not Transmissible in
Mammals 124
6.6 Risk of an Outbreak Escaping Local Control of Pathogens that Are Transmissible in Mammals 126
6.6.1 Effect of enhanced transmissibility in mammals on risk of an outbreak escaping local
control 128
6.6.2 Effect of enhanced pathogenicity on risk of an outbreak escaping local control 134
6.6.3 Effect of overcoming/evading natural/residual/innate immunity on the probability of an
outbreak escaping local control 135
6.6.4 Effect of loss of containment pathways on risk of loss of local control of an outbreak 136
6.7 Consequences of a Global Pandemic of Pathogens that Are Transmissible in Mammals 139
6.7.1 Seasonal Influenza Virus 139
6.7.2 Pandemic Influenza Virus 142

6.7.3 Avian Influenza Virus That is Transmissible Amongst People 143
6.7.5 Effect of enhanced transmissibility in mammals on consequences of a global outbreak 145
6.7.6 Effect of enhanced pathogenicity on consequences of a global outbreak 149
6.7.7 Effect of countermeasures evasion on consequences of a global outbreak 153
6.7.8 Effect of evasion of natural/residual immunity on consequences of a global outbreak 159
6.8 Supporting an Estimate of Absolute Risk 161
6.9 Using the Parametric Risk Assessment: Example Calculation 165

6.9.1 Step 1: Determine if the probability of the pathogen escaping the laboratory changes 165
6.9.2 Step 2: Determine the change in the probability of a resulting outbreak escaping local
control 165
6.9.3 Step 3: Determine if the consequences of a resulting pandemic changes 166
6.9.4 Putting it together 167

6.1 Overview of Results
The Biosafety Risk Assessment evaluated the increase in risk to human health of pandemics caused by
modified strains of the influenza viruses and the coronaviruses. In every case, the increase in risk
compared to wild type strains was provided to determine if GoF experiments could create pathogens that
are more likely to cause laboratory acquired infections, more likely to create a local outbreak or more
likely to cause a global outbreak (or cause one of greater consequence) than those strains that evolved via
natural forces. Note that although this study identified several types of risky, theoretical GoF experiments,
many of these experiments have not been described in the literature. For example, no examples of
researchers endeavoring to determine if seasonal influenza viruses could be made more transmissible
were found. As another example, if a virus grows to a conveniently high titer naturally (e.g. 1E8 pfu/ml)
then enhancing this level of growth may not be desirable or, indeed, biologically feasible. Moreover,
many GoF studies are performed in highly attenuated strains, so that even though the risk of an outbreak
increases if these strains were modified, risk is increasing from a very low level.
GoF Phenotype Seasonal Influenza Pandemic Avian Influenza Coronaviruses

Viruses Influenza Viruses Viruses
Increases Increases Increases
probability of an probability of an probability of a
Enhanced
outbreak and the outbreak and the global outbreak and
transmissibility
consequences of an consequences of an consequences of a
outbreak outbreak global outbreak
Enhanced Increases Increases
pathogenicity consequences consequences
Decreases
Adaptation to
N/A N/A probability of an N/A
mammals
outbreak
Increased
Evasion of induced consequences in
N/A
immunity high income
countries only
Increases
Evasion of probability of an
natural/residual outbreak and the N/A N/A
immunity consequences of an
outbreak
Increased Increased
consequences in consequences in
Antiviral resistance N/A
high income high income
countries only countries only
Enhanced growth in Increased chance of Increased chance of
culture/eggs a LAI a LAI
Figure 6.1. A figure showing increase in risk of research on GoF pathogens over wild type pathogens. The
darker the shade of gray, the more a GoF phenotype increases risk of human illnesses and deaths. Marked in
white are GoF phenotypes that are not relevant (N/A) to risk or reduce risk.
In short, a strain of influenza virus that is as transmissible (or to which the population has as little
minimal immunity) as newly emerged pandemic strains WHILE leading to a case fatality rate of more
than 5%, would pose more of a risk of a global pandemic than any wild type strain heretofore identified.

No experiments that are likely to be conducted under the rubric of GoF research will drive risk more than
this combination of traits or significantly increase the risk of a laboratory acquired infection. All other
combinations of traits would lead to pathogens that have a lesser total risk than the wild type 1918 strain.
Increasing the transmissibility of the coronaviruses, while significantly increasing the risk of work with
those pathogens, still creates a pathogen that poses less of a risk of a global pandemic than the wild type
1918 influenza strain.
In the brief section that follows, we provide the rationale behind these overall conclusions by showing
how changes in each GoF phenotype affects each component of the risk assessment for each pathogen.
Further supporting evidence is provided in this Chapter in Sec. 6.4 and onward.
6.1.1 Seasonal Influenza Viruses
This risk assessment appropriately considers the fact that not all loss of containment events lead to a
laboratory acquired infection, that not all laboratory acquired infections initiate a local outbreak (because
infected workers may be given prophylaxis or be isolated) and that not all local outbreaks initiate a global
pandemic. In fact, at each step, only a minority of events initiate the next step. Figure 6.2 shows the
probability of each step in the chain of events that would eventually lead to a global pandemic from a loss
of containment incident for wild type seasonal influenza assuming that the previous step has occured.
From this figure, only 2% of laboratory acquired infections, which are rare in themselves (but not
quantified by our method), start a local outbreak (that is, cause at least one secondary case) and only 20%
of local outbreaks would seed a global pandemic. Moreover, in the case of seasonal influenza, the risk of
a global pandemic is exacerbated by laboratory research only if that laboratory is working on a strain that
has not circulated recently because residual immunity is likely to curtail its spread. If the strain is
currently in circulation, the spread of the natural outbreak is likely to be driven by travelers, not by
laboratory accidents.
Figure 6.2. Relative probability of each step in the event chain from a loss of containment event to a global
pandemic for a loss of containment event involving seasonal influenza.

To understand how GoF research could influence risk from research on seasonal influenza, it is useful to
consider each step in the incident that leads from a loss of containment event to a global outbreak and to
comprehend how the GoF trait would influence either probability or consequences at each step. Figure 6.3
divides risk by each step in the biosafety RA and shows how GoF research influences risk for seasonal
influenza.
GoF Phenotype Increase Increase Increase Probability Increasing Global

Probability of Probability of an Outbreak Escapes Consequences
Lab Acquired a Local Local Control
Infection Outbreak
Enhanced
N/A Less than 2x 2-3x 2x
transmissibility
Unknown—possible
Enhanced pathogenicity N/A N/A 10x or more
decrease in risk
Adaptation to mammals N/A N/A N/A N/A
Evasion of induced 4x in the high income
Less than 2x Less than 2x N/A
immunity countries only
Evasion of
natural/residual Less than 2x Less than 2x 2-3x 3-4x
immunity
5x in the high income
Antiviral resistance Less than 2x Less than 2x Unknown
countries only
Enhanced growth in
2x* N/A N/A N/A
culture/eggs
Figure 6.3. A figure showing increase in risk of GoF research on seasonal influenza over wild type seasonal
influenza. The darker the shade of gray, the more a GoF phenotype increases risk of human illnesses and
deaths. Marked in white are GoF phenotypes that are not relevant to risk (N/A), reduce risk or that could not
be quantified. The star denotes a result that may not be statistically significant.
Because seasonal influenza viruses are associated with a low case fatality rate, increasing this rate could
significantly increase the global death toll from an outbreak, increasing risk. Developing seasonal
influenza strains that are more transmissible than wild type strains (approximately as transmissible as
pandemic strains) or that overcome residual immunity increase the probability that an outbreak would
escape local control and increase the consequences should a global outbreak be initiated. The creation of
an antiviral resistant strain could increase the consequences of a global outbreak, but only in high income
countries where caches of these antivirals could be handed out to a significant fraction of the infected
population.
A strain of seasonal influenza that can overcome protective vaccination could also increase the
consequences of an outbreak in high income countries, which has the resources to vaccinate their
populations quickly. However, this phenotype is of concern only if it enables the virus to evade the
protection afforded by means other than changing its antigenic properties, which is not a subject of
current research in influenza. (The vaccines made in response to an outbreak caused by a laboratory
accident would be raised specifically for the strain causing the outbreak, so it would “match” the novel

antigenic properties of the strain.) The relatively small increase in risk due to vaccine resistance is not
unexpected due to the modest efficacy of seasonal influenza vaccine at preventing infection.336
6.1.2 Pandemic Influenza Viruses
Figure 6.4 shows the relative probability of each step in the chain of events that would eventually lead to
a global pandemic from a loss of containment incident for pandemic influenza. From this figure, fewer
than 1% of high-risk loss of containment events, which are rare in themselves but not quantified here,
involving wild-type pandemic influenza would lead to a laboratory associated infection, only 5% of
laboratory acquired infections start a local outbreak and only 30% of local outbreaks seed a global
pandemic.
Figure 6.4. Relative probability of each step in the event chain from a loss of containment event to a global
pandemic for a loss of containment event involving pandemic influenza.
Figure 6.5 divides risk by each step in the biosafety RA and shows how GoF research influences risk for
pandemic influenza. The risk analysis suggests that the only GoF research that creates a strain of
pandemic influenza that poses more risk of a global outbreak than a wild type strain (in this case, the 1918
pandemic strain) is a strain that is antiviral resistant, and the increase is modest. That being said, even this
trait increases the consequences in high income countries only because poor countries lack access to a
sufficient quantity of antivirals to effectively protect their populations. GoF manipulations in other strains
of pandemic influenza could increase their risk of causing a global outbreak, but would still match the risk
posed by the wild type 1918 pandemic strain. Similarly, enhancing viral growth in culture beyond that
which is achievable in wild type strains (1E9 or 1E10/ml) increases the probability that a laboratory
acquired infection would occur (by five- or 15-fold, respectively). However, it is doubtful if this
phenotype is desirable or scientifically achievable because growth to 10E8 is sufficient for almost all
purposes except the production of vaccines (using attenuated strains).
336 Cowling, BJ et al “Assessment of influenza vaccine effectiveness in a sentinel surveillance network 2010-13, United
States.” Vaccine 2015, article in press.

GoF Phenotype Increase Increase Increase Increasing
Probability of Lab Probability of a Probability an Global
Acquired Infection Local Outbreak Outbreak Escapes Consequences
Local Control
Enhanced
N/A Less than 2x Less than 2x Less than 2x
transmissibility
Enhanced Unknown—possible
N/A N/A Less than 2x
pathogenicity decrease in risk
Adaptation to
N/A N/A N/A N/A
mammals
Evasion of induced
Less than 2x Less than 2x N/A Less than 2x
immunity
Evasion of
natural/residual Less than 2x Less than 2x Less than 2x Less than 2x
immunity
2-3x increase in
Antiviral resistance Less than 2x Less than 2x Unknown high income
countries only
Enhanced growth in
Up to 6x N/A N/A N/A
culture/eggs
Figure 6.5. A figure showing increase in risk of GoF research on pandemic influenza over wild type pandemic
deaths. Marked in white are GoF phenotypes that are not relevant to risk (N/A), reduce risk or that could not
be quantified.
6.1.3 Avian Influenza Viruses
Wild type avian influenza is insufficiently transmissible amongst people to cause a global outbreak driven
by the spread of the disease among humans. For this reason, no loss of containment event would lead to a
global outbreak from a wild type strain.
avian influenza. Because wild type strains of avian influenza cannot spread globally between people, the
creation of strains that are human transmissible would greatly increase the risk that such an outbreak
could occur, which could cause millions of illnesses. The creation of a strain that is as transmissible as
seasonal influenza would have a significant chance of sparking a global outbreak if a local outbreak were
initiated. An increase in the pathogenicity in humans of the most pathogenic, wild-type strains increases
the consequences modestly only if one assumes that the case fatality rates of the most pathogenic strains
of avian influenza are inflated by the underreporting of mild illness in people. Adapting avian strains to
humans without increasing transmissibility (thereby lowering the median infectious dose in people)
actually decreases risk because, while it increases in the probability that a single laboratory worker would
become infected, it decreases the risk that birds would become infected through an accidental release via
the solid waste stream, which otherwise could lead to thousands of human infections and is the dominant
loss of containment pathway. No other GoF trait increases the risk posed by avian influenza.

GoF Phenotype Increase Probability Increase Probability Increase Probability Increasing Global
of Lab Acquired of a Local Outbreak an Outbreak Consequences
Infection Escapes Local
Control
Wild type strains
Wild type strains Wild type strains
Enhanced cannot cause a
Less than 2x cannot cause a local cannot escape local
transmissibility global outbreak
outbreak control
of human disease
Enhanced
N/A N/A N/A Less than 2x
pathogenicity
Adaptation to
Decrease in risk N/A N/A N/A
mammals
Evasion of induced
immunity
Evasion of
natural/residual N/A N/A N/A N/A
immunity
Antiviral
resistance
Enhanced growth
Less than 2x N/A N/A N/A
in culture/eggs
Figure 6.6. A figure showing increase in risk of GoF research on avian influenza over wild type avian
deaths. A numerical value cannot be provided for the greatest increases because the risk from wild type
pathogens is vanishingly low for these outcomes. Marked in white are GoF phenotypes that are not relevant
to risk (N/A) or reduce risk.
6.1.4 Coronaviruses
the coronaviruses. Recall that the RA uses the word “coronavirus” to mean the coronaviruses that cause
SARS or MERS and not the coronaviruses that cause the common cold. Importantly, most estimates of
the transmissibility of the coronaviruses consider these pathogens to be insufficiently transmissible and
sufficiently susceptible to control measures such that a global pandemic has a very minimal chance of
occurring. Even using the highest estimates of R0 for SARS-CoV, which are derived from the location
and time of an outbreak that spread the most, results in less than a 10% chance of sparking a global
pandemic should a local outbreak begin. For this reason, increasing the transmissibility of the
coronaviruses could significantly increase the chance of a global pandemic due to a laboratory accident.
That being said, even if these strains were modified to be as transmissible as pandemic influenza, the
viruses’ long generation time and lack of asymptomatic transmission, which results in susceptibility to
control measures, the resulting outbreaks would still be contained a majority of the times they were
initiated. Increasing the pathogenicity of these strains could also increase risk somewhat through the
increase in global deaths expected, especially since most deaths from wild type strains are suffered by
those with significant co-morbidities. Strains of the coronaviruses that have enhanced growth properties
could increase risk of a laboratory acquired infection if samples with 1E9pfu/ml or 1E10pfu/ml were
routinely manipulated in a laboratory (risk would increase by seven- or 25-fold, respectively). However, it

is uncertain if this phenotype is desirable or even achievable because wild type coronaviruses grow to a
sufficiently high titer for manipulations in the laboratory.
GoF Phenotype Increase Increase Increase Probability Increasing

Probability of Lab Probability of a an Outbreak Escapes Global
Acquired Infection Local Outbreak Local Control Consequences
Wild type strains do
Enhanced Less than 2x Several orders
N/A not escape local
transmissibility increase of magnitude
control
Enhanced
N/A N/A 2-3x increase
pathogenicity
Adaptation to
N/A N/A N/A N/A
mammals
Evasion of induced
N/A N/A N/A N/A
immunity
Evasion of
natural/residual N/A N/A N/A N/A
immunity
Antiviral resistance N/A N/A N/A N/A
Up to a 10-fold
Enhanced growth in increase in
N/A N/A N/A
culture/eggs probability for a
1E10pfu/ml culture
Figure 6.7. A figure showing increase in risk of GoF research on avian influenza over wild type
coronaviruses. The darker the shade of gray, the more a GoF phenotype increases risk of human illnesses and
deaths. A numerical value cannot be provided for the greatest increases because the risk from wild type
pathogens is vanishingly low for these outcomes. Marked in white are GoF phenotypes that are not relevant
to risk (N/A).
However, if a coronavirus were modified such that it caused a global pandemic (one in which sustained
human-based transmission occurs in all global regions, which has never been observed), their relatively
long incubation time and disease course (compared to influenza) lead to a pandemic that unfolds over
many years (Figure 6.8). While some outbreaks peak within two years, most require two to ten years to
reach their peak. The fact that the outbreak evolves slowly gives public health authorities more time to
adapt and expand their efforts to further contain the outbreak than the modeling conducted in this
assessment suggests.

Figure 6.8. The number of coronavirus outbreaks modeled that peak (in terms of new cases per day) at any
particular day after the global outbreak begins. To show the duration of truly global outbreaks, outbreaks
that lead to less than one million infections are not shown.
6.2 Methodology
6.2.1 Purpose of This Task
The purpose of the quantitative biosafety RA is to provide information regarding the risk (in terms of
consequences and probability) of a release of strains of pathogens with novel phenotypes that enhance
their pandemic potential due to an accident or natural disaster. This assessment has three main
components:
1. The estimate of the probability that an accident/natural disaster would occur and result in an
infection of a human or other animal outside the laboratory,
2. The estimate of the probability that an outbreak that occurs would escape beyond local control
and seed a global outbreak, and
3. The estimate of the extent of an outbreak that would result from an infection outside the
laboratory.
Critically, because GoF research occurs in a world in which research on dangerous, wild type pathogens
is ongoing, the risk assessment is comparative. That is, we seek to determine how much risk increases if
GoF research proceeds.

6.2.2 Input from Modeling Subject Matter Experts
To guide our modeling effort, we interviewed the following infectious disease modeling subject matter
experts (SMEs): Dr. Jason Asher, Dr. Steven Riley, Dr. Martin Meltzer and Dr. Carrie Manore.337,338,339,340
Their input is reflected in the modeling methodology described in this section. All of the interviewed
experts unanimously agreed that the use of multiple models, covering event initiation, initial local spread,
and potential global outbreak, respectively, is reasonable and sound. Additionally, all experts confirmed
that the choices of a stochastic approach for the initiation phase of an outbreak followed by a homogenous
mixing, deterministic approach for modeling the global spread phase were appropriate. Mr. Asher spoke
about the BARDA Interactive Flu Model, an SEIR-type model, and confirmed that it contained the
necessary features for the biosafety risk analysis; this model became the basis for global outbreak
simulations.
When asked about appropriate stochastic models for the initiation phase of the outbreak, Dr. Riley
suggested that, in lieu of a computationally-intensive agent based model, a branching process model may
be more appropriate. He believed that such an approach would capture the key features of such an
outbreak, while leaving out dimensions that were not critical in determining whether an outbreak would
grow beyond local control, such as where infected individuals lived. He also remarked that branching
process models would capture a critical facet of laboratory acquired infections: that most of them do not
lead to outbreaks of significant size. Moreover, an agent-based model would require the parameterization
of features of the environment of the outbreak that would be unknowable. Dr. Riley emphasized the
criticality of the shape of the offspring distribution in such a model and suggested we speak with Dr.
James Lloyd-Smith, with whom Gryphon collaborated and consulted on the development of the
branching process model used in the final risk analysis.341 All other interviewees to whom a branching
process model was mentioned either raised no objections or confirmed the appropriateness of the
approach.
In searches of the literature, little data and few models covering zoonotic infections of influenza were
found. Dr. Manore agreed that relatively little data existed, particularly for interspecies contact rates, and
that few people were considering models that incorporated humans, domestic animals, and wildlife in a
single model. She remarked that their approach to overcoming this lack of data in their influenza models
was to parameterize based on a retrospective analysis of prior outbreaks to ensure that the predictions of
the model were reasonable. This approach is clearly not suitable for a prospective analysis such as ours.
6.2.3 Interplay of the Components of the Biosafety Risk Assessment
The biosafety RA has several components that will be married together to understand how risk of a
laboratory accident changes if GoF experiments proceed. These components and their interplay are
shown in Figure 6.9.
337 Leidos contract support to the Division of Analytic Decision Support, Biomedical Advanced Research and Development
Authority, Department of Health and Human Services, Washington, United States
338 MRC Centre for Outbreak Analysis and Disease Modelling, Department of Infectious Disease Epidemiology, School of
Public Health, Imperial College London, United Kingdom
339 National Center for Emerging and Zoonotic Infectious Diseases, Centers for Disease Control, Atlanta, GA, United States
340 Center for Computational Science, Tulane University, New Orleans, LA, United States
341 Department of Ecology and Evolutionary Biology, University of California, Los Angeles, California, United States of
America

Figure 6.9. Flow diagram showing the interplay of the various components that permit the calculation of the
probability and severity of an outbreak to calculate relative risk. The components in blue speak to the
probability that an infection outside the laboratory occurs. The components in orange speak to the
probability that an outbreak will escape local control and cause a global pandemic. The components in green
speak to the consequences of a global outbreak. Quantitative modeling approaches are shown in bold italics.
All components are considered together to understand probability and consequence using a Probabilistic
Risk Assessment framework.
To inform the RA, we interviewed 77 Subject Matter Experts as shown in Figure 6.10. These stakeholders
provided data on the frequency of GoF experiments and the experimental conditions, containment
features, health surveillance procedures, isolation procedures and public health response measures that
occur in their laboratories.

Risk Analysis: SME Interviews
9
4 Clinician
32 Epidemiological modeling
Laboratory biosafety
Local/state public health
25
PPP research
7
Figure 6.10. A pie graph showing the sector from which the 77 SMEs who informed the RA were drawn.
6.2.4 Taking a Parametric Approach to the Analysis
Perhaps the most challenging aspect of this effort is to assess the risk of experiments that have yet to
occur in locations that haven’t been identified performed under unknown biosafety conditions. If there
were a finite number of phenotypic changes to the pathogens of interest, a finite number of experimental
procedures used, a small number of possible locations for the research or a finite number of conditions
under which the research could occur, we could simply test each one. However, this approach is not
feasible and would not provide the NSABB with the information it needs to make recommendations and
for the USG to formulate policy.
Worse still, drawing a "bright-line" boundary between research that qualifies as GoF and research that
does not is likely to be difficult. Even for experiments of the same type, the specific strains selected, the
quantities used, and the intended outcomes matter greatly: the context of the research is key. As a result,
without a method that allows an examination of the details of each experiment and the pathogens therein,
accurately assessing the relative risks of each would difficult if not impossible.
For this reason, the power of the modeling approach was exploited to explore the entire space describing
each quality of GoF and wild type pathogens used, the experiments undertaken, the laboratories and the
containment measures where the experiments occur. This type of approach is called “parametric analysis”
because each quality, or parameter, is allowed to vary across a range of possible values. This approach
enabled not only the accurate assessment of the risk of future experiments, but also can be used to support
the development of generalizable guidelines after important drivers of risk are identified.
This approach was applied to the pathogens themselves, for which the following parameters were allowed
to vary, including:
• Pathogenicity in humans (including case fatality rate and infectious dose),

• Transmissibility in humans,
• Evasion of diagnostics and countermeasures, including vaccines and anti-virals,
• Evasion of immunity, either natural or induced, and
• Growth in culture or eggs

The approach was applied to the experimental conditions, for which the following parameters were
allowed to vary, including:
• The number and type of infected animals,

• The frequency animals are physically handled,
• The concentration and volumes of stocks and infected tissue culture samples, and
Although we considered pathogens with a range of characteristics, any of these pathogens were
considered to have arisen from an already established strain of virus in terms of the experiments
conducted and epidemiology. Moreover, we specifically parameterized and modeled wild type influenza
viruses (seasonal, pandemic and avian), MERS Co-V, SARS Co-V (to establish the baseline for relative
risk) and the strains that have arisen from GoF experiments already (to identify any change in risk).
This approach was applied to the laboratories that may perform the work, for which the following
parameters were allowed to vary, including (a full list is available in the Supplemental Information):
• The existence and effectiveness of various containment features,

• The existence and effectiveness of various pieces of personal protective equipment (PPE),
• The existence and effectiveness of decontamination procedures,
• The existence and effectiveness of monitoring systems for the health of the workforce, and
• The population density of humans outside the laboratory,
To understand risk of GoF work when performed under less-than-ideal circumstances, as may be the case
in other parts of the world, we can assess how the removal of any particular containment feature,
decontamination procedure or health monitoring procedure would affect the probability of a release.
6.2.4.1 Bounding the Parametric Approach in Science
The parametric analysis described above is also underpinned by scientific data. That is, a parameter could
be allowed to vary between any arbitrarily large or small value but is grounded by information related to
the attributes of known pathogens. For example, the contagiousness (which could be measured in the
number of naive people an infected person can infect) could be allowed to vary from zero to the entire
population of the earth. However, given data on real viruses, we know that this parameter value would
rarely exceed ten, and only when describing the most contagious viruses known. Moreover, the possibility
that a modified avian influenza virus becomes more contagious than any influenza virus that has ever
been observed stretches reason. Similarly, even though the parametric analysis will allow the systematic
removal of containment features, it is highly unlikely that GoF experiments would proceed without any
containment whatsoever. As another example, an extremely risky strain of influenza virus may be one
that could be simultaneously transmissible in poultry and humans, however such a strain may not be
possible due to the nature of the changes in viral receptors that lead to changes in species tropism.
Our parametric analysis allowed us to evaluate how risk changes as the GoF research features change.
The possibility that risk increases significantly only when a parameter reaches an unrealistic value builds
confidence in the fact that our model is capturing all possible facets of risk. At the same time, this
counterfactual analysis provides some information as to which manipulations or settings are unlikely to
pose a significant increase in risk over work with unmodified pathogens.
Although the intent was to develop an RBA approach that is flexible enough to encompass all possibly
risky manipulations of any pathogen, many aspects of a pathogen’s lifecycle (and its epidemiology) are
unique and therefore models that faithfully replicate the risk of outbreak initiation and spread must use

real examples. Moreover, the experimental conditions used in GoF experiments (volumes, titers, cell
lines, animal models, etc) are also unique to the pathogens used in the experiments. Due to the focus of
GoF concerns on influenza and coronaviruses, we used these viruses as the basis of our work in this
study.
6.2.5 Probability of an Infection Outside of Containment
6.2.5.1 Choosing the incidents to model in detail
To estimate the probability that an accident or natural disaster (together called incidents) leads to an
infection outside the laboratory, we treated each separately. There are several types of incidents that could
cause a loss of containment and a subsequent outbreak outside a containment laboratory. To identify the
incidents to model, we leveraged previous laboratory risk studies and reports on past incidents to
understand which incidents most drive risk of outbreaks caused by incidents at containment laboratories.
Minor accidents, which do not drive risk because they were found to be unlikely to cause an infection or
to have minimal consequences should they occur, need not be considered in detail. Recall that because
risk is the product of the consequence and frequency of an adverse event, the riskiest accidents to examine
include a variety of types: 1) those that are frequent and low-consequence, 2) those that are rare and high-
consequence, and 3) those that are not uncommon and of moderate consequence. After the identification
of the incidents that drive risk, the remainder of the biosafety RA analysis focuses on the evidence basis
and modeling of only these most risky accidents. We further winnowed out incidents that were found to
be minimally risky in the context of GoF research.
We collected past laboratory RAs and Environmental Impact Assessments (listed in Appendix III). For all
studies that quantitatively assess risk in terms of probability and consequence, we identified the highest
risk incidents and gathered all data related to those incidents. For studies that simply detail scenarios that
are deemed to be “maximum reasonably foreseeable events” or a “plausible, worst case” scenario, we
determined if quantitative studies examined similar incidents and where they fall on the risk ranking. As
these types of scenarios are typically chosen because they have maximal consequences, without a
consideration of probability of the event occurring, it is possible that these so-called “maximum
reasonably foreseeable events” or “plausible, worst-case” scenarios are so vanishingly unlikely (i.e.
occurring less than once in a billion years) that they do not affect risk much, even though they are
consequential when they occur. If these scenarios were found to be relatively low risk, they were
excluded from further analysis. If they are were assessed in other quantitative risk assessments or there
was no other reason to exclude them, they were included in our Fault Tree Analysis. This process
explicitly captures the low-probability (but plausible), high-consequence events.
To supplement our list of high-risk accidents from previous assessments, we examined accident reports
and case studies (sources are listed in Appendix III). Importantly, historical incidents are supported by a
minimal amount of quantitative information (mostly related to consequence) that prohibits an estimate of
risk. Just because an accident occurred once, we cannot calculate the probability that it would happen
again. For this reason, we compared the list of historical accidents to the list of incidents to model from
past RAs. Also, we found that many incident reports are included as high-risk incidents in past RAs (for
instance, “spill”) and other incidents are components of an overall sequence of failures that leads to a
release in a past RA. For example, “PPE failure”, which is mentioned in accident reports, is a possible
failure node in all incidents that involve the generation aerosols or splashes on personnel. We included
these types of events in several fault trees to assess the influence of the failure of these systems on risk of
another incident (such as a spill). From these riskiest accidents, we removed those accidents that do not
apply to the pathogens we are considering. For example, the National Bio- and Agro-defense Facility Site
Specific Risk Assessment conducted for the Department of Homeland Security in 2012 study identifies

several risky accidents arising from the fact that their pathogens are studied in large animals (like cattle),
which can physically break containment features.
The list from previous studies and reports consists of the riskiest incidents that cannot be discounted from
previous studies, the most common accidents that could lead to an infection outside a laboratory, any
accident that did lead to an infection outside the laboratory (that cannot be discounted) and all “maximum
reasonably foreseeable events” that could not be shown to be lower in risk than incidents included.
Even though the highest risk accidents are unlikely to change much in their frequency regardless of the
nature of the pathogen, the probability that an infection outside the laboratory could occur may be
significantly different. That is, although the chance that a centrifuge rotor breaks is the same if the sample
inside contains a virus or a bacteria, the chance that an infection may occur outside the laboratory if a
worker caries infected material out on his shoe may be different if the contaminant were Foot and Mouth
Disease virus vs influenza virus. For this reason, we determined if the pathogens of particular interest in
GoF studies, specifically influenza viruses, MERS-CoV and SARS-CoV pose unique risk pathways that
must be investigated further due to dissimilarities of their biology, pathogenesis, host range or life cycle
compared to the pathogens considered in past RAs. From this qualitative analysis, we identified further
accidents to consider to capture the unique risks that these pathogens may pose. Specifically, animal bites
were a “low risk” incident in past RAs, but ferrets, an animal model of choice in influenza studies, are
particularly prone to biting (which, although the risk of infection from the bite is unlikely for respiratory
pathogens, could deposit pathogen directly on the skin, increasing the risk of self-innoculation into the
eye or nose). An “incident” modeling the fact that infected animals are constantly exhaling pathogen
(called “animal respiration”) was also specifically included because, unlike in other laboratories, infected
animals pose a direct hazard to unprotected workers (should containment fail). All highest risk, relevant
incidents from past studies and case reports were combined with these additional selected incidents to
define the list of high-risk incidents that we were investigated in detail.
6.2.5.2 Predicting the Amounts and Pathways of Pathogen Releases for Accidents
To assess the probability that an accident would occur resulting in a loss of containment, we used Fault
Tree Analysis (FTA), an accident modeling approach in which each possible system component that
could fail in a complex pathway to an accident is explicitly parameterized with probability and
consequences of failure. We implemented this FTA using Monte Carlo simulations, an approach that
randomly samples values from all possible parameter values to explore the effect of uncertainty on our
analysis. This approach was used to determine the probability that an accident occurs while an experiment
with a dangerous pathogen is taking place (or in the handling or shipment of a pathogen) and the amount
of material that escapes containment. Should an accident occur, the accident will have consequences in
terms of the dispersion of the material. The dispersed material will then be subject to elimination or
retention due to decontamination procedures and containment systems. The fact that many accidents may
not be apparent when they occur is important to consider because additional measures are usually
implemented when overt accidents occur. The accident could generate an infectious aerosol, fomites or
living carriers (laboratory animals or workers). We also considered that the possibility that an accident
generates many types of sources (a centrifuge spill could create an aerosol, fomites and infected workers).
For accidents, the frequency of experiments and the concentration of the stocks and samples manipulated
were estimated to describe the “opportunity space” for accidents to befall. These data were gathered in
site visits and interviews with PPP researchers. Once an accident occurs, the agent may be released but
will still be inside of containment. The effectiveness of containment measures determines how much
material leaves containment depending on the nature of the accident. Containment measures reduce the
concentration of a biological release, but may not be in place/functioning where the accident occurs due
to human error (e.g. mislabeling/mishandling of a sample) or equipment failure. Data on the failure of

mechanical systems and of human error rates are derived from the scientific/engineering literature. Data
supporting the Fault Tree Analysis are provided in the Supplemental Information along with full
descriptions of the Fault Trees. Given the role of humans in historical laboratory accidents, our FTA
includes a robust consideration of human reliability in the execution of appropriate decontamination and
safety procedures.
Any material that escapes containment and decontamination described a source term that is used to model
the initiation of an infection outside the laboratory. Aerosols were described by their quantity in a
respirable range. Fomites were described by their material, location and quantity. Infected people and
animals were described by their type and quantity. Insofar as an accident causes consequences inside the
laboratory to the workers, these casualties were tallied as consequences.
6.2.5.3 Source Terms for Natural Disasters
For natural disasters, we estimated, at any given time, how much pathogenic material is in the laboratory
that could be released. This pathogenic material could be in the form of stocks in storage, samples being
manipulated or infected animals. The disaster itself may lead to several events inside the laboratory (the
spill of materials or the release of animals) and the disruption of containment systems (overpressuring of
HEPA systems or breach/failure of a building envelope). Several infection pathways could
simultaneously lead to outside infection after a natural disaster (an earthquake could lead to the
generation of an agent aerosol and the escape of infected animals). FTA was used to determine the
probability that a natural disaster occurred and affected pathogen stocks, infected animals or experiments
in progress. As described above, we examined only those natural disasters that are deemed to be high risk.
The dispersed material will then be subject to elimination or retention due to decontamination procedures
and containment systems, although these may be compromised due to the disaster. Natural disasters
cannot be covert and so we assume that special public health measures (such as social distancing or
restrictions on movement) would be implemented if a natural disaster is known to strike a containment
laboratory.
Any material that escapes containment and decontamination helps describe a source term that was
modeled for its ability to cause an infection outside of a laboratory. The source terms were described
similarly to those arising from accidents.
6.2.5.4 Modeling Initiation of an Infection outside a Laboratory
Once infected material leaves the laboratory, it may cause infections in nearby human or animal
populations. The probability of the infection occurring depends on the nature of the source term, which
can be aerosols, fomites or infected animals/researchers. Each type of source term was modeled using a
separate methodology.
Indoor and outdoor source terms were modeled separately. Indoor source terms were modeled as if the
worker causing the accident inhaled all of the aerosol to understand a maximum level of risk. We chose
this approach because a worker who creates an aerosol is exposed to a relatively high concentration of the
contaminant until it disperses within the room. Workers who did not create the aerosol are exposed by
dwelling in a completely well-mixed space that is slowly exhausted to the outside. Any material escaping
the building was modeled using the Hazard Prediction and Analysis Capability (HPAC), a model
developed by the Defense Threat Reduction Agency, which is able to predict the transport and downwind
infections over large areas given real population densities. We chose two laboratory locations to
understand the range of risk from aerosols, New York City—the urban area with the greatest population
density in the US—and a small town. Real weather data for those locations was used. One hundred
releases were modeled ranging over a variety of times of year and times during the work day. HPAC was

used to calculate the dose that people downwind received. Dose/ response curves were used to determine
how much of the population inhaling the pathogen becomes infected. Given data on the populations of
susceptible animals (specifically ducks) and their minute tidal volume (the amount of air they inhale per
minute), we calculated animal infections over the same area.
For infected animals that leave the laboratory due to a natural disaster, we presumed that an outbreak is
initiated by the animal encountering a human before it expires (for human-transmissible pathogens) or by
encountering a susceptible bird (for bird-transmissible pathogens). For infected animals that leave the
laboratory because they are carried out intentionally in a malicious act (relevant to the biosecurity RA,
below), we presumed that the malicious actors are themselves infected (for human-transmissible
pathogens) or that the infected animal encounters a susceptible bird (for bird-transmissible pathogens).
For the animal escape incident in the biosafety RA, our FTA models predict the animal to leaving the
laboratory is vanishingly unlikely (by bolting, unnoticed through several self-closing doors) and instead
drives risk by escaping containment features within the lab and infecting workers.
For infected workers, we created a separate FTA that accounts for their behavior, the possible violation of
health monitoring procedures and isolation guidance and their contacts with susceptible individuals
throughout their disease course. Some protocols are initiated only if the exposure event was overt and
considered high risk (an observed spill for example). Other protocols, such as the reporting of influenza-
like illness and isolation should such symptoms appear, occur regardless of the type of accident that
caused the illness. Workers may violate protocols (via ignorance or arrogance), and these workers enter
into the models of local infection, as described below. Also, a worker could initiate a local outbreak if
they develop no clinical symptoms or develop transmissible illness prior to the onset of symptoms.
For fomites, we developed a stochastic, Markov chain model to predict the likelihood of an outbreak
initiating after a laboratory worker leaves containment with virus on his or her person. The model tracks
the contamination through the paths it must take to result in infection of the initial laboratorian, of one or
more household or community members, or of avian species on a farm (or any combination of the three).
All infections are the result of internalization of the virus from a contaminated surface or body part; that
is, this is a model of contamination transference and subsequent infection, not a model of contagious
transmission. The transference model utilizes Monte Carlo simulations to estimate the likelihoods of a
number of possible actions that would lead to internalization, spread, or removal of the virus. Human
infection occurs when viral contamination on a person’s hand enters their mucosal membranes of the eye,
nose, or mouth, and the probability of infection is dose-dependent based on the calculated amount of virus
present at the time of inoculation. For an animal infection to occur, the primary laboratorian must
encounter a susceptible species, at which point it is assumed that all of the virus is inoculated into the
animal. For animal contact to occur, the worker may need to violate quarantine protocol, which occurs at
a specific probability, after which visits to an animal facility occur at a predetermined rate, as with the
events above.

Figure 6.11. Schematic of the transference and infection model.
For the data supporting the parameter values used in the models that support the estimate of source terms
causing infections outside the laboratory, please see Supporting Information.
This analysis predicts the probability that an outbreak would be initiated by human or avian infections
outside the laboratory for each of the possible incidents modeled. These incidents can vary by the species
infected (ducks or people), the type of person infected (laboratory worker or a member of the public) and
the number of people infected.
6.2.6 Predicting the probability that an outbreak escapes local control
Once a human-transmissible disease leaves the laboratory and infects at least one person outside of
containment, an outbreak is initiated (recall that health monitoring and isolation/quarantine measures
enacted for exposed laboratory workers are already considered before the outbreak occurs as described
above). Depending on the release, an outbreak can start with one or more initial cases. For example, a
large aerosol release from a catastrophic incident (like an earthquake) could infect many dozens of
people. We considered outbreaks that initiate with the infection of a laboratory worker differently than
outbreaks that begin with a member of the public because we presume that laboratory workers (and their
families) would be more likely to report to public health authorities if they developed unusual symptoms
of infectious disease and would be more likely to self-isolate.
An outbreak that starts with a handful of people is governed by stochastic forces that could, by chance,
cause the outbreak to extinguish. Similarly, an outbreak that is recognized early and subjected to vigorous
control measures may extinguish.
To model the local outbreak, we used a branching process model, developed by, and in consultation with,
Dr. James Lloyd-Smith (UCLA), a recognized world expert in stochastic epidemic modeling.342
Branching process models are stochastic, where each case creates a number of new cases based on a
probability distribution. In the model used in this report, the distribution is a negative binomial
distribution with parameters R0 (the average number of new cases each case generates) and k (which
342 J. O. Lloyd-Smith, S. J. Schreiber, P. E. Kopp & W. M. Getz. Superspreading and the effect of individual variation on
disease emergence. Nature 438: 355-359 (2005)

reflects the variation in infectiousness between individuals, where low values of k imply high variation
and high values of k imply low variation). Low k is appropriate for MERS/SARS because most people
create no secondary cases, whereas some create a very large number. Higher k is appropriate for flu
because many people infect one or two others, some zero, some a large number. A discussion of the
choice for the values of R0 and k for wild type strains of influenza and the coronavirus infections is
provided in the Supplemental Information. Branching process models capture one crucial feature of new
outbreaks: that many new outbreaks extinguish at a small number of cases.
The branching process model we adapted considers various control measures and can account for partial
immunity in the community (important for outbreaks of recently circulating influenza strains). Social
distancing and isolation/quarantine are parameterized. Because our analysis is not to evaluate control
measures but to compare the risk of various outbreaks, we explore a variety of plausible values for the
parameters describing these measures. The parameter values that describe control measures are described
in the Supplemental Information. Notably, our model tracks laboratory workers and community members
separately, so that we subjected each to different control measures.
In our analysis, an outbreak was considered to be out of control if either of the following conditions were
met:
• The model calculated that, given the number of cases in the current generation, that the outbreak
had less than a 5% chance of extinguishing at any point in the future, or
• That any generation included more than 1,000 infected individuals (which probably outstrips the
ability of a locality to control), or
• The model includes 200 generations of infected individuals without extinguishing or reaching any
other termination condition (suggestive of never getting under control).
In this Risk Assessment, 2.6 billion simulations were performed in our BPM to provide statistically sound
data to explore the parameter set for wild type and GoF pathogens and a variety of outbreak control
parameters.
Once an outbreak was considered out of control, it was considered to seed outbreaks globally. The
illnesses and deaths due to an outbreak that extinguishes either due to stochastic forces or due to control
measures were tallied as part of the consequences of the local outbreak.
6.2.7 Modeling the Global Consequences of a Human-Transmissible Outbreak
Once an outbreak was found to grow out of local control using the branching process model, we modeled
the global consequences of a pandemic using the HHS-BARDA Interactive Influenza Model (IIM), which
is used by the Centers for Disease Control and Prevention and HHS-BARDA to evaluate the effectiveness
of medical countermeasure strategies to control influenza outbreaks.343 IIM is a “Susceptible, Exposed,
Infectious, Recovered” (SEIR)-based model, which is a compartmental epidemiological model which
tracks the progression through various stages of a disease course of individuals in an outbreak. IIM
considers the differences in vaccination and clinical visit rates of different age groups (children, adults,
and the elderly), contact rates between these groups and control measures, like mass vaccination, social
distancing and anti-viral treatment.
343 For example, see Biggerstaff, M et al “Estimating the potential effects of a vaccine program against an
emerging influenza pandemic--United States.” Clin Inf Dis S1, S20-9 (2015).

IIM was developed using contract rates and demographic data for the US. To globalize the model, we
collected demographic data for 12 regions of the world, divided by geography and income (with the
rationale that high-income countries have distinct demographics and public health resources than other
countries). We characterized each region by population, class size (used to scale school-based contact
rates), household size (used to scale household contact rates) and age stratification (used to scale relative
numbers of children and the elderly). The methodology for scaling contact rates is described in Appendix
III Sec.14.3.1 and the demographic data supporting the regionalization of the globe is provided in the
Supplemental Information. Although this method captures some demographic differences between
regions of the world, it does not capture cultural practices and socioeconomic factors (like underlying
poor health) that could affect the outbreak. Also, public health measures, like social distancing, are
assumed to be equally effective in all parts of the world (however, vaccine doses and antivirals are more
limited).
If an outbreak escaped local control, we assumed that it would continue to seed infections in the US and a
US-wide outbreak will continue to seed outbreaks abroad. For this reason, travel rates were unnecessary
to obtain as eventually the disease would spread. Each region was seeded with 100 initial cases.
Parameter values used in the IIM model are provided in the Supplemental Information.
To support the analysis in this Risk Assessment and adequately explore the parameter space, the IIM ran
approximately 750,000 simulations.
6.2.8 Simplified Modeling of Bird-Transmissible Pathogens
One hypothetical consequence of a laboratory release of research with a strain of influenza that is
transmissible only amongst avians is that a strain could establish itself in wild bird populations (by
infection via an aerosol or contaminated worker or contaminated waste leaving the laboratory), causing
sporadic human disease over a dispersed geographic area, similar to the natural H5N1 strain today. For
this eventuality to occur following loss-of-containment and subsequent release, a series of events must
occur: the virus must reach an environment where infection of a wild bird can occur; it must infect a wild
bird; the virus must spread and migrate with a population of birds; these infected wild birds must then
spread the virus to a domestic bird population; this virus must then spread from the domestic birds to
humans, and finally, the virus need be capable of causing disease in a human host. Note that a virus that
spreads between humans is presumed to spread between humans efficiently and any incidental
transmission from birds will not significantly affect the kinetics of the outbreak; hence, this section does
not consider human transmissible viruses. This presumption is supported by the opinion of several of the
interviewed experts, who believed that a gain of function influenza virus, including the H5N1 strains
adapted to transmit between ferrets by the airborne route, could be adapted to spread efficiently among
humans or among birds, but not between them due to differences in viral receptors in these animals. This
belief agrees with the historical evidence, as we have yet to identify either a natural human influenza that
spreads easily among birds or a natural avian adapted virus with sustained mammalian transmissibility.
6.2.8.1 Unpredictability of the consequences of novel avian-influenza strains
Determining the probability and consequences of each of the events necessary for an avian virus to infect
humans is very difficult primarily due to missing data. For example, one reference reviewed 4,763
literature sources of human to animal transmission of any disease, and found no documented examples of
direct human to animal transmission of influenza.344 Similarly, despite detection of influenza in natural
344 Messenger AM et al (2014) Reverse zoonotic disease transmission (zooanthroponosis): a systematic review of seldom-
documented human biological threats to animals. PloS one 9: e89055

water sources and measurements of the persistence of influenza in water, suggesting that “cloacal
drinking” by waterfowl of contaminated water may be a source of infection, no identified source listed an
ID50 for such a process.345,346,347 However, of the sources of uncertainty, estimating the consequences to
humans of a flu circulating in wild birds remains the largest due to uncertainties in the biology of the
virus and the role of human-avian interaction in its epidemiology.
Despite intense research efforts spanning decades, predicting the transmissibility and pathogenicity of a
new or novel avian strain in humans or other mammalian hosts remains challenging. Part of this difficulty
stems from the diverse range of symptoms and effects seemingly similar strains cause, combined with the
apparent uncorrelated symptom severity between birds and mammals. Shown in the Supplemental
Information are data for eleven recent avian influenza outbreaks, eight of which have caused human
cases. Comparing the outbreaks reveals the unpredictability of human effects. For example, despite the
virus that caused the 2015 H5N2 outbreak containing a hemagglutinin (HA) in the same clade as one
known to cause fatal human H5N1 infections, the H5N2 outbreak has of yet caused no known human
cases of infection.348,349 This difference could be behavioral (due to enhanced biosafety practices in the
poultry industry in the USA), or may be due to differences in biology of the strains. The H7N7 outbreak
of 2003 caused only one human fatality and most symptoms were restricted to conjunctivitis, even though
the strain appeared highly infectious to humans, with 250/500 of potentially exposed humans tested
showing evidence of seroconversion.350 In comparison, the ongoing H7N9 outbreak causes minor to no
signs in either wild birds or poultry, but causes severe respiratory disease in humans in the relatively few
human cases it has caused.351
The distribution of an outbreak is as unpredictable as its transmissibility and pathogenicity. The majority
of poultry outbreaks of influenza remain constrained to one or a few flocks, with a few spreading much
further. The current outbreak of H5N1 began in December 2003 with the first reported human cases in
Vietnam and spread rapidly.352 By April 2004 it had spread to Thailand, Korea, Japan, Indonesia and
Hong Kong and by November 2004 to mainland China. By February 2006 it had become intercontinental,
spreading to Europe as well as Africa, where it remains endemic to Egypt. The timing and location of
spread appeared to correlate with bird migratory patterns, hinting at wild bird-mediated spread.353,354 In
contrast, H7N9 began in the same global region, and appeared to initially spread more quickly, yet despite
beginning in the same region and presumably being subject to the same cultural and geographic factors, it
has only spread through a geographically contiguous area and not spread internationally, confounding
determination of whether the spread is primarily wild bird or human mediated.355 Meanwhile, the North
345 Deboosere N et al (2011) Development and validation of a concentration method for the detection of influenza a viruses
from large volumes of surface water. Applied and environmental microbiology 77: 3802-3808
346 Stallknecht DE et al (1990) Persistence of avian influenza viruses in water. Avian diseases 34: 406-411
347 Alexander DJ (2007) An overview of the epidemiology of avian influenza. Vaccine 25: 5637-5644
348 Ip HS et al (2015) Novel Eurasian highly pathogenic avian influenza A H5 viruses in wild birds, Washington, USA, 2014.
Emerging infectious diseases 21: 886-890
349 de Vries E et al ibid.Rapid Emergence of Highly Pathogenic Avian Influenza Subtypes from a Subtype H5N1
Hemagglutinin Variant. 842-846
350 Fouchier RA et al (2004) Avian influenza A virus (H7N7) associated with human conjunctivitis and a fatal case of acute
respiratory distress syndrome. Proceedings of the National Academy of Sciences of the United States of America 101: 1356-
1361
351 World Health Organization, “Overview of the emergence and characteristics of the avian influenza A(H7N9) virus”, Report
issued May 31, 2013.
352 Yee KS et al (2009) Epidemiology of H5N1 avian influenza. Comparative immunology, microbiology and infectious
diseases 32: 325-340
353 Liang L et al (2010) Combining spatial-temporal and phylogenetic analysis approaches for improved understanding on
global H5N1 transmission. PloS one 5: e13575
354 Gilbert M et al (2006) Anatidae migration in the western Palearctic and spread of highly pathogenic avian influenza H5NI
virus. Emerging infectious diseases 12: 1650-1656
355 Bui C et al (2015) A Systematic Review of the Comparative Epidemiology of Avian and Human Influenza A H5N1 and
H7N9 - Lessons and Unanswered Questions. Transboundary and emerging diseases

American H5N2 outbreak began in the Pacific Northwest and quickly jumped two migratory flyways to
the Midwest; the mechanism for this rapid eastward spread has not yet been identified.
The lack of a solid scientific evidence basis for predictive epidemiology in avian influenza viruses implies
that any serious quantitative analysis would be unfounded. For this reason, we took a simplistic approach
to modeling outbreaks of influenza viruses that spread between birds only.
6.2.8.2 Spread of escaped laboratory virus to wild or domestic birds
First, avian-influenza strains are modeled in the Fault Tree Analysis like any other strains. We have
enough data to predict the chance of infection of a human or a bird when exposed to a source of pathogen.
We can therefore quantitatively predict if humans or animals are infected within the laboratory (due to a
variety of incidents) or outside the laboratory (due to aerosols or transfer of contamination from a worker
to poultry). Should a bird be infected outside the laboratory or an infected bird escape from the laboratory
(in the earthquake and biosecurity scenarios), we presume that an avian influenza outbreak occurs and has
consequences similar to the recent outbreaks. That is, we presume that between 0 and 1,000 human
infections occur and that the case fatality rate is between 0 and 50%. Given the lack of data, our model
presumes an equal probability of any result in this range. Because we are not estimating economic
consequences or risks to animal health, this approach is sufficient to characterize the risk of this agent to
humans given the paucity of data available.
Recall that if the pathogen is transmissible between people (regardless of if the strain is a natural one or if
it is a modified avian-influenza strain), we modeled the outbreak assuming that all human health risk is
dominated by human-to-human contact.
6.2.9 Estimating Risk of Experiments Involving GoF Pathogens
Each modeling component is used to predict a single aspect of risk:
1. The Fault Tree Analysis is used to determine how pathogen characteristics, containment features,
experimental manipulations and the laboratory environment contributes the probability of escape
and the number of cases that would initiate an outbreak (a component of consequence),
2. The branching process model estimates the probability that a local outbreak would grow and
seed a global pandemic. If the outbreak extinguishes due to stochastic factors or due to an
effective public health response, the consequences from the local outbreak are tallied, and
3. The HHS-BARDA Interactive Influenza Model is used to predict the global consequences of a
pandemic.
By linking the outputs of the modeling component, we can state how much any pathogen, research feature
or environment drives risk. For example, we can ask the specific question of how much does fatality risk
change if one increased the transmissibility of H5N1 influenza in humans to half of that of seasonal
influenza. We would explore how the probability and consequences of all laboratory accidents depends
on this change, and how the probability of a local outbreak escaping control depends on this change and
how the consequences of a global outbreak depends on this change. Comparing the three modeling
components together provides an overall estimate of the change of risk.
In fact, for each pathogen phenotype and condition under which GoF research could be performed, we
determined how varying the phenotype or research condition within scientifically defensible limits

influences risk. To undertake this sensitivity analysis, we determined how risk changes if a parameter
value is held to a series of specific values in a Monte Carlo analysis in which other parameter values are
allowed to be selected at random. This analysis determined if risk increases or decreases as any specific
parameter value changes across the range of possible values for all other parameters.
Each parameter that is found to significantly influence risk (either positively or negatively) compared to a
baseline that assumes work with unmodified pathogens was further explored to understand the reason
behind this relationship. In this way we determined if a parameter value must be set to an unreasonable
value in order to significantly drive risk or if risk can be increased at parameter values that could be easily
expected. Moreover, by analysis of the results, we determined if risk is driven only when a combination
of parameter values occurs (risk increases significantly only if the pathogenicity and transmissibility of an
agent is increased, or only if transmissibility is increased and the work is performed without worker
health monitoring). Together these results will help identify the GoF activities and conditions that could
significantly increase risk of an outbreak compared to work with wild type pathogens.
The branching process model and the IIM are computationally intensive and so a Monte Carlo analysis
could not be done to explore the entire parameter space. Instead, a variety of parameter values were
rationally chosen to defining the epidemiology (e.g., R0 and latent period) of the viruses and the efficacy
of control measures to contain the outbreak. For each of these parameters, a range of values believed to
cover a significant fraction of the possible parameter values were used, and results were obtained for each
unique combination of every parameter varied. Figure 6.12 illustrates the variation in simulation results
for these parameters for seasonal flu global outbreaks, where each marker represents a result for a unique
combination of parameters.
Figure 6.12. Illustration of results for single models in simulations. Each circle represents the results from a
single model run with a single set of parameters. Lines connect models that differ only in the R0 value used in
the simulation. To reduce the number of lines for visual clarity, some parameters were fixed: shown are
models using the median values of antiviral efficacy on mortality, case fatality rate, and fraction of cases
symptomatic, as well as no community mitigation; all other parameters ranged across their values
appropriate for seasonal influenza.
Elsewhere in this report, when showing results for these simulations, the median number of deaths across
all values of the varied parameters is shown by the marker, and the 10th and 90th percentile value of deaths
across the same parameters are shown by vertical lines extending outward from those markers (the “error
lines”). These vertical lines do not represent statistical uncertainty in the underlying simulations; instead
they represent uncertainty as to the properties of the virus, outbreak and public health capacities. If a real-

world outbreak were to occur with a defined set of parameters matching those simulated by one of the
models, the results would, with high confidence, match those of that particular model. In addition, while
these vertical lines plotted cover 80% of the resultant number of deaths for the model parameters
simulated, they should not be understood as a typical 80% confidence interval. Because no probability
distribution was assigned to the underlying varying parameters, the vertical lines represent the middle
80% of the outputs of the simulations, and not the range of the 80% most likely outbreaks. This approach
suggests that a real outbreak would follow the overall shape of the median line presented (and not simply
move randomly inside the range presented), but may be higher or lower (up to the bounds of the “error
bars”) if certain properties of the virus or the control measures caused the outbreak to spread more
aggressively (or less aggressively) than the median set of parameters.
6.3 Practices in GoF Laboratories that Reduce Risk but Are Not Included in Our Study
To collect data to inform the modeling approach, interviews were conducted with laboratorians, biosafety
officials and public health officials. These interviews uncovered several measures that certainly reduce
the risk posed by containment research but in ways that could not be included in a quantitative study that
models human behavior abstractly. This section describes some of these practices that speak well of the
culture of biosafety that exists in these containment laboratories but could not be captured by our models.
A thorough examination of current practices in influenza and coronavirus biosafety level 3 (BSL-3) was
conducted through site visits and interviews with researchers, public health officials, and institutional
representatives. Best practices in biosafety and biosecurity pertaining to gain-of-function research were
identified that exceed recommendations or requirements from various bodies, including the Occupational
Safety and Health Administration (OSHA), select agent regulations, recommendations of the Federal
Experts Security Advisory Panel (FESAP), and Institutional Animal Care and Use Committees (IACUC).
Practices either unique to specific institutions or commonly found across institutions are highlighted and
were found to be especially beneficial/optimal/useful in training, exercises and drills, laboratory practices,
health precautions, physical security, and institutional culture.
6.3.1 Training
Laboratory directors and personnel at various BSL-3 laboratories shared their protocols for training new
researchers. From this, several best practices are highlighted. One observed prerequisite to BSL-3 work is
demonstrating competency in BSL-2 work. Additionally, across all institutions, extensive BSL-3 training
was observed, involving both written examinations and supervision of hands-on laboratory skills. One
institution described a tiered training structure, in which the first tier covered basic laboratory operations,
emergency situations, and general laboratory safety. The second tier covered more specific training for
laboratory safety when performing cell culture work. The third tier covered procedures and precautions
for animal work. Each tier was associated with a training checklist, which a trainer would use to assess
the trainee. Another institution required the trainee to shadow the trainer in the BSL-3 laboratory and
perform laboratory procedures under mentored supervision before conducting independent work. Best
practices for hands-on training involve dedicated one-on-one instruction and active roleplay for scenarios
such as an animal bite or a biohazardous spill.
Training and education can be codified into standard operating procedures (SOPs) covering experimental
protocols, biohazardous spills, working with animals, potential exposure to infectious material, and
biosecurity threats. These SOPs can be made easily accessible within the BSL-3 containment lab, should
the need arise. Demonstrating knowledge of all SOPs can be required as part of BSL-3 training.
Additional training in biosecurity is also recommended, covering topics such as cybersecurity, identifying
abnormal or suspicious behavior, identifying insider and outsider threats, and how to deal with strangers

requesting lab access. Institutions remarked on the need for constant reminders and renewal of training to
counter complacency. Commendably, some institutions were particularly thorough about BSL-3 training
requirements. Visiting researchers were required to repeat BSL-3 training, even if they had prior
experience either elsewhere or at that same institution. Training is not only limited to issues of biosafety
and biosecurity. One institution provided communications training for researchers on how to discuss gain-
of-function work in public settings. Finally, institutions can offer select agent training to first responders
to inform what agents may be present during an incident and what to do in case of a large-scale spill or a
fire.
6.3.2 Exercises and Drills
Hands-on training can extend beyond laboratory protocols to tabletop exercises and drills within and
outside the laboratory setting. Institutions described several exercises and drills, such as responding to a
researcher having a medical emergency in the BSL-3 containment lab, responding to a potential exposure
in the laboratory, and responding to a natural disaster. Another research facility discussed methods for
testing their security infrastructure, such as leaving a door open or holding up signs to security cameras to
test for prompt response. On a wider scale, the research institution can conduct exercises and drills in
conjunction with first responders, environmental health and safety (EHS), and local hospitals for better
preparation against a potential exposure. Examples of such exercises are: a researcher following SOPs for
exposure to a pathogen, a researcher not following SOPs for exposure and showing up at a hospital
emergency room, and response to a bomb threat. Conducting these drills also strengthens cross-
institutional relationships, which can better inform future preparedness and response protocols. One
institution asked local first responders to perform walkthroughs of the research facility to learn how to
gain access and what to do during an emergency. For instance, the fire department was instructed to
contain but not extinguish a fire in the BSL-3 containment lab, allowing it to burn within those
boundaries. Notably, one institution remarked that whenever a researcher would display influenza-like
illness, this essentially became an exercise in practicing SOPs for a potential exposure. Finally, a best
practice that formalizes these relationships is to establish an Emergency Operations Center (EOCs) under
the parent institution or university to better coordinate emergency responders, EHS, local public health,
and the research facility. EOCs can run campus-wide drills to scenarios such as bomb threats, natural
disasters, and active shooters to prepare a coordinated response effort from multiple agencies.
6.3.3 Laboratory Practices
CDC select agent regulations dictate several requirements for day-to-day laboratory operations, including
a regular inventory of pathogen stocks and inspections of laboratory equipment and the BSL-3 facility.
Several institutions have demonstrated particularly useful laboratory practices that may surpass regulatory
requirements or otherwise represent optimal biosafety and biosecurity measures in access control,
inventory, animal work, facility maintenance, and communications. Furthermore, select agent
requirements represent best practices for non-select agent labs that work with, for instance, seasonal
influenza viruses.
Several best practices in access control are highlighted here. One non-select agent status laboratory was
observed to keep its freezer containing pathogen stocks under lock and key and to perform frequent
inventory checks. Another practice was to grant access to select agent freezers only to a small number of
staff out of the many more who were approved for BSL-3 work. This can prevent researchers from
performing unauthorized experiments, as it requires explicit permission to access the pathogen stocks.
Another institution required researchers to obtain permission to access anesthetic drugs for anesthetizing
animals for in vivo work. More broadly speaking, it would be a best practice to control access to reagents
necessary for risky experimental protocols.

Maintaining an updated inventory is a requirement for laboratories working with select agents. However,
one institution was noted to count inventory more frequently compared to peer institutions (monthly
instead of quarterly). Alternatively, another facility performed random inventory checks. One institution
randomly sampled 10% of its boxes to reconcile its contents with the inventory log. If discrepancies were
noticed, a 100% inventory check was performed, and the CDC was notified. Another select agent
requirement is to limit how long experimental samples may be kept. The best practice observed for this
requirement was to keep experimental samples up to 30 days, after which they were either discarded or
added to the permanent inventory. There are several additional best practices associated with counting
inventory. One institution required two people present to count inventory. One researcher was
“permanent” and was always present at every inventory check. The other researcher was “rotating” to
witness the inventory count and ensure that inventory was not simply memorized as a complacent way of
counting. Another institution assigned one employee to keep track of all changes to inventory; this
employee was responsible for conducting counts and was to be notified if a sample were to be taken from
stocks.
Additional practices were noted that improve the safeguarding of inventory. Witnesses can be required for
any changes to inventory, including taking agents from pathogen stock, destroying old samples, and
adding samples. Stocks not used for at least one year can be archived in boxes sealed with security tape.
Researchers highlighted several best practices when working with animals in the course of pathogenic
research. One is to limit researchers’ and animal caretakers’ contact with laboratory animals, a USDA
regulation though not a CDC regulation. Animals can be observed prior to the conduct of experiments to
determine whether they are prone to abnormal or aggressive behaviors, which may make them more
likely to inflict bites or scratches on their handlers. These behaviors will be noted for experimenters so
they can take appropriate precautions when working with those animals. Furthermore, animals can be
completely or partially anesthetized before experimental procedures to prevent bites or scratches. One
research group noted that genetically mixed mice were more prone to aggressive behaviors and thus
partially anesthetized all mixed mice as a precaution. A daily check, including weekends and holidays, of
animals and other laboratory equipment can be conducted. In order to record which employees were
trained to perform animal experiments, animal husbandry, and respiratory testing, one facility kept an
animal handling training sheet. Finally, a paper trail for each laboratory animal can be maintained, which
details its history of procedures, tests, and bodyweight measurements, as well as the dosage and strain of
the experimentally induced infection.
Briefly, some best practices were noted with regards to maintaining the facility and its equipment.
Frequent inspections of the shower and facilities can ensure that containment safeguards and
decontamination procedures remain optimal. Additionally, several facilities performed annual shutdowns
for several weeks in order to perform a comprehensive surface and gas decontamination and to perform
preventive maintenance.
There were several practices observed that sought to optimize researcher-to-researcher communications
or to utilize a partner system to limit mistakes or malicious behavior. A radio system can be used to
communicate between BSL-3 researchers and outside staff. One institution mandates that any potential
exposure, no matter how minor and even if it does not breach PPE or skin, should be reported over the
radio. This allows an outside employee to be aware of the situation, and furthermore the employee can
guide the BSL-3 researcher on next steps, preventing a possibly stressed researcher from making rash
decisions. Another simple tool is to place a whiteboard outside the BSL-3 containment lab that displays
which researchers are working in which suites, and which pathogens are in each suite. One best practice
that was especially notable was notification of weekend or after-hours work. Researchers seeking to
conduct work off-hours can be asked to notify a coworker by phone of time of entry, expected duration,

and time of exit. One institution employed an on-call cell phone, which is always kept on and is assigned
to an employee by rotation. Messaging this phone is required for after-hours work in the laboratory.
Finally, several institutions require BSL-3 laboratory staff to wear emergency “man-down” pendants,
which can used in the event of an emergency to alert first responders and research supervisors.
With regards to a partner or two-person system, when interviewing different research institutions,
different opinions emerged on its utility. Many institutions required the partner system when performing
experiments requiring animals or sharps. Some institutions used the partner system liberally, requiring
witnesses to validate changes to inventory (as mentioned above) or proper execution of inactivation
protocols (inactivating an agent to transfer from BSL-3 to BSL-2). However, institutions differed in their
opinions about the partner system when performing more routine experiments. One institution encouraged
the partner system whenever possible. However, researchers at another remarked that the risk of
accidental exposure was higher with two people, and that the two-person system provided little utility. It
is important to point out that the utility of the two-person system has historically been contentious, and
that no applied research has been done to assess the benefit of such a measure.
Lastly, some additional best practices for day-to-day laboratory conduct are to limit a researcher’s hours
in a BSL-3 lab to 3-4 hours daily and to designate one employee to receive and sign off on shipped
biological materials. These can limit the chance of exposure and ensure an extra degree of security,
respectively.
6.3.4 Health Precautions
Several best practices were identified that better reduce the chances of severe illness following a
laboratory exposure. These can be categorized as conditions of employment, post-exposure SOPs, and
partnerships with local and state public health departments and with local hospitals.
Some institutions were observed to require employees who worked in BSL-3 laboratories to abide by
certain rules. One commonly observed best practice was the requirement of the seasonal influenza
vaccination as a precaution against laboratory-acquired influenza infection. Another was to medically
clear new employees, which would (1) discover any underlying medical issues that may exclude a
researcher from working with select agents, and (2) obtain a baseline serum sample prior to starting lab
work, in order to test for seroconversion in the event of potential exposure. Lastly, one institution
obtained signed statements from its employees agreeing to self-quarantine, self-report body temperature,
permit home visits by a nurse, and submit samples for diagnostic testing, in the event of a potential
exposure.
One best practice for post-exposure SOPs is to include extra precautions following a potential exposure.
For instance, one institution isolates the exposed researcher and administers an N95 mask without an
exhalation valve while awaiting emergency medical response, even though the pathogen would not be
expected to replicate within those few hours following exposure.
Partnerships between the research institution, local and state public health, and local hospitals can be
established prior to an exposure incident to expedite the medical response. Researchers can carry cards
describing their occupation and what agents they work with, which should be shown at the emergency
department to facilitate proper treatment. Medical emergency protocols for laboratory pathogens used in
the neighboring research institute can be shared with the local emergency department, and occupational
health concerns for working with these pathogens can inform hospital protocols for safety and security. In
fact, this can be further codified into a memorandum of understanding (MOU) with the local hospital. It
was noted that if the institution is a university, hospital physicians can often be affiliated with the

university’s medical school, which facilitates a culture of cooperation between the hospital and research
staff. The contributions of institutional culture to best practices in biosafety and biosecurity are explored
in a later section. Finally, one institution has shared samples and genomic sequences with the state public
health department to verify that their diagnostic tests detect the virus strains commonly used in the
laboratory, in the event of a potential laboratory-acquired infection.
For employees leaving the university, terminate access 2 weeks prior to leaving and must go through exit
physical before they leave, to ensure they’re not sick. (2 weeks based on incubation period of
SARS/MERS – 10 days.) In some laboratories, everyone must check into lab daily. If someone doesn’t
show up, the lab is responsible for tracking them down. Lab will notify EHS if they are unaware of
someone’s whereabouts, and EHS will reach out to the university hospital ER to let them know to watch
out for that person to show up.
6.3.5 Institutional Culture
Several researchers cited their institutional culture as a powerful factor in promoting safety and security in
the laboratory. Institutional culture can dictate workplace satisfaction, willingness to report incidents,
awareness, and workforce turnover, all of which can directly or indirectly influence the levels of biosafety
and biosecurity. Several institutions noted the importance of developing a non-punitive culture that
encouraged over-reporting, especially of “gray-area” incidents such as a minor spill without breach of
PPE or skin. Also widely practiced was a culture of carefulness and vigilance, bolstered by consistent
reminders to practice good safety and security measures to prevent complacency. One institution
remarked that the principal investigator sets the example by obtaining all biosafety and biosecurity
training. Many institutions cited their small work environment as conducive to maintaining vigilant
security, since all of the staff knew each other. One supervisor commented about developing an intuition
for the happiness levels of all staff members, which can reduce the risk of an insider threat. Institutions
can additionally offer assistance programs for employees to cope with hardships or obtain counseling.
Establishing an environment that promotes staff retention is also a best practice. This builds relationships
between laboratory staff and is a strong security measure in limiting the number of new employees.
Additionally, one institution does not allow undergraduate students to work in the BSL-3 lab, due to
concerns with turnover and with the length of time needed for BSL-3 training.
Strong support from the parent institution for the gain-of-function research program can also promote a
positive working culture. One research group noted that, in the face of controversy, the parent institution
remained strongly supportive of the research program, which encourages the laboratory staff to be
diligent about reporting incidents and maintaining a safe and secure working environment.
Finally, as mentioned above, strong relationships between the research institute and local hospitals, first
responders, regional FBI offices, EHS, and local and state public health departments contribute to a
positive institutional culture that lends itself to better preparation for and response to laboratory incidents.
6.3.6 Additional Institutional Policies
Finally, additional best practices were observed in various institutions that do not fall under the above
general categories. Several institutions required principal investigators to register their research with EHS,
documenting a notice of intent (listing the agent of study, purpose of the study, and dual use research
questions), a risk assessment, and training requirements. As part of this registration, EHS can perform an
annual inspection of the facilities to verify the proper safety and security measures. Institutions have also
employed campus-wide behavioral risk assessments to monitor for behaviors or emails of concern.

Finally, institutions can share their own practices with other research facilities, improving each other’s
security and safety procedures.
6.4 Probability of Laboratory Acquired Infections
6.4.1 The selection of incidents to include in the RBA
In this study, we analyzed 10 previous laboratory accident risk assessments, and three compilations of
accident/incident reports to identify the accident or incident scenarios that would be quantitatively
evaluated in our study. Any scenario that was high risk (either due to their frequency or consequence) or
used as the “maximum reasonably foreseeable events” in at least one source was included for quantitative
analysis, except when:
• General accident types that are explored in more detail by another accident type (e.g. “waste
stream” would be discarded in favor of the high risk “leaking pipe” scenario),
• Any incident with an unknown cause because these are not quantifiable (e.g. “contamination
outside laboratory with unknown cause”—note that these are likely captured by other event
types), and/or
• Accidents specific to containment research on large animals
Other scenarios were considered but not included in a quantitative analysis because they are rarer than
events that would have a larger consequence. Beyond these events, we included additional scenarios to
capture risks that may inhere in GoF research specifically or were recently in the news, specifically:
• Floods, due to the flooding of hospitals and laboratories that occurred during Hurricane Sandy,
and
• Animal bites because of extensive work with ferrets, which tend to bite more often than mice or
guinea pigs
In many cases, the “incidents” identified in other reports aren’t incidents in themselves but risk factors
that influence the risk of other incidents. That is, if the HVAC system fails, this failure has a consequence
only if animals are actively exhaling pathogen into the ambient air or there is a spill or splash. For these
events, their probability of occurrence was included in ALL other relevant accident fault trees. In total, 16
incidents were investigated in detail to form the basis of our quantitative analysis.

Table 6.1. Rationale for Scenarios included in Risk Benefit Assessment.
Scenario Rationale for Inclusion
Splash incident Recognized as high risk in the NBAF
Spill incident Recognized as high risk in NBAF
Failure to keep containment in place Recognized as high consequence in NEIDL. Included as a factor in other incidents
Solid waste incident Recognized as high risk in NBAF
HVAC failure HVAC failure was noted in reports and could be of potential high consequence. Included as a factor in other incidents
Equipment-- failure of containment feature Common in incident reports. Included as a factor in other incidents
Equipment--power loss at facility NEIDL estimates high risk and actual examples in reports
Many examples in reports, human error (with equipment failure, human failure to check inactivation (e.g., recent anthrax
Improper inactivation of pathogen
at DoD)). Investigated separately but also is part of other incidents investigated (waste streams and splashes).
Transference--glove to skin due to improper removal High risk in NBAF. Included as a factor in other incidents
Shipping accident Considered exceptionally high risk in NBAF; examples in reports of accidents (though of no consequence)
Recognized as high risk in the NEIDL while other RAs state low risk; some examples in literature of animals escaping or
Animal--escape from containment
otherwise disappearing
Improper inactivation of liquid waste Recognized as high risk in NEIDL
Centrifuge release Canonical high risk scenario in almost every RA
Natural Disaster--earthquake Most catastrophic scenario for NBAF and NEIDL
PPE Failure Several examples in literature leading to actual laboratory acquired infection
Protocol failure--use of wrong containment Several examples in literature leading to actual LAI. Included as a factor in other incidents
Puncture is the most common cause of reportable lab accidents. Because the GoF pathogens are probably not infectious
Puncture/sharp object injury via injection, this incident is considered to lead to a breach in the gloves that creates a contamination on the hands (leading
to possible later inoculation of the worker or a contact).
Waste-liquid waste leak/pipe burst High risk in NEIDL
Animal--bite/scratch Although low risk in NEIDL, may be higher frequency in ferrets
Exhaled pathogen escapes laboratory (animal
Due to contagious nature of some GoF pathogens, this scenarios deserves quantitative evaluation
respiration)
Natural Disaster--flood Low risk but recent examples present - Galveston and New York
Scenarios listed in blue are failure modes that could exacerbate the risk of a loss of containment and are included in other events. Scenarios listed in green were included by name
in this assessment.

A.
B.
C.
Figure 6.13. Pie charts showing the types of incidents included in our study and the fraction of total risk or
incidents they comprise in previous studies or reports. Only the incidents extracted from the pie chart are
excluded from further analysis in this RBA. A. The NEIDL, B. The NBAF, C. past reports.

Note that these events include the low probability, high consequence incidents; the high probability, high
consequence incidents and the high probability, low consequence incidents from other reports, along with
the “maximum reasonably foreseeable events” incidents. Excluded from this point on in the assessment
are only the events that are rare and inconsequential, or not foreseeable (for instance, have a probability of
less than the age of the earth). Note, any event that is reasonably foreseeable (even if extremely unlikely)
and of high consequence was captured in our assessment. These 16 incidents capture the vast majority of
the risk from previous assessments: more than 99% of the risk from the NIEDL study, about 90% of the
risk from the NBAF study (much of the rest relates to work with large animals only), and 80% of incident
reports (the rest are not quantifiable due to unknown cause) (Figure 6.13 above).
All incidents included in our study were studied to reconstruct the pathways leading to the loss of
containment event. After study, some incidents were excluded from quantitative analysis because no
plausible scenario that leads to a loss of containment could be identified to model. Other incidents were
quantitatively investigated but excluded from the fault tree analysis because all pathways identified lead
to vanishingly unlikely and/or small releases. See Appendix III for details.
6.4.2 Identification of locations at risk of earthquakes and floods
To quantitatively assess the risk of earthquakes and floods at GoF laboratories, we found the GPS
coordinates of 36 containment laboratories, including labs that were formerly conducting GoF research
and several additional BSL-4 and BSL-3 facilities that are currently operational or under construction.
The flood risk at each location was assessed using information from the Federal Emergency Management
Agency.356 The earthquake risk was assessed using information from the US Geological Survey. 357 Of
these 36 locations we identified the locations of greatest risk of flooding and earthquake and assessed the
risk of a loss of containment event at that facility due to a natural disaster. If the risk was significant, we
would have assessed the risk from these natural disasters at other sites with slightly lesser risk. However,
we found that the risk of natural disasters was minor compared to accidents, so this analysis was not
performed.
6.4.3 Irreducible uncertainty prevents an accurate prediction of absolute risk
Humans are an integral component of every laboratory, however, humans are prone to making mistakes
due to carelessness, haste, tiredness or unfamiliarity with validated procedures. In most complex systems,
the physical systems (fans, valves, filters, alarms) are demonstrated to fail much less often than the
humans operating the systems and interpreting the alarms these systems make when an error occurs. The
only human reliability data found directly related to work in a containment laboratory are studies of
decontamination (when removing gloves or washing hands). Much of the data on human reliability comes
from the transportation, chemical and nuclear sectors and this study had to analogize to interpret human
error rates to laboratory situations. Because of the absence of data, in this risk assessment some
conservative assumptions were made that prevent the accurate estimation of absolute risk. None of these
assumptions affect the relative risk of an accident with a modified pathogen compared to a wild type and
so the comparative risk assessment still holds.
Conservative assumptions made include:
356 FEMA, National Flood Hazard Layer Map (Official), accessed in June, 2015 at
http://fema.maps.arcgis.com/home/webmap/viewer.html?webmap=cbe088e7c8704464aa0fc34eb99e7f30.
357 USGS, US Seismic Design Maps, accessed in June, 2015 at http://earthquake.usgs.gov/designmaps/us/application.php

• When a skill error (a slip) happens with a sharp object (scissors, typically) during a necropsy, the
assumption is that the slip results in a cut through the worker’s glove(s). There are no data on the
relative number of errors during necropsy that result in damage to the specimen, dropping of the
instrument or any other inconsequential failure compared to breaches in the gloves, therefore this
assumption was made to conservatively maximize risk,
• When a splash happens when working with pathogen (for example a contaminated pipette tip
skipping over the top of a well), the splash is assumed to land on the worker’s hands and not the
hood or any part of the clothes or body unlikely to contact the worker’s face or others outside the
laboratory. There are no data on distribution of drops from laboratory accidents on this scale so
the assumption of contamination on the gloves was made to conservatively maximize risk,
• When a worker contaminates their hands by any pathway, the contamination is assumed to be on
the fingertips because this part of the hand is mostly likely to contact a contaminated surface.
This is a conservative assumption because the fingertips are the only part of the hand to permit a
self-inoculation (in the eye or nose) to maximize risk,
• When gloves fail, they are assumed to fail on the fingertips because these parts of gloves are the
most prone to failure. Note that this assumption forces the point of contamination and glove
failure to be coincident, which maximizes risk,
• When an accident the worker directly caused leads to the generation of an aerosol (like the spill
of a viral stock), the assumption is made that the worker inhales all of the aerosolized pathogen
because they are nearest to the most concentrated part of the spill (assuming that the aerosol
reaches equilibrium in the room does not account for the fact that the worker was relatively close
to the source).
6.4.4 Relative Probability of Laboratory Acquired Infections
The approach used in this risk assessment predicts that a variety of accidents, when combined with human
or equipment failures, lead to laboratory acquired infections from work with the GoF pathogens. For
seasonal influenza, most laboratory acquired infections are the result of aerosols accidentally generated by
spills or centrifuge accidents, while a minority are caused by contamination of the hands during necropsy,
cell culture or via an animal bite. For pandemic influenza, because of the additional respiratory protection
used under BSL3 conditions, events that contaminate the hands cause slightly less than half of the
laboratory acquired infections, while the rest are caused by aerosols. In avian influenza laboratories, the
vast majority of infections are those of wild birds contaminated by the accidental discharge of
incompletely decontaminated solid waste. Less than 10% of the accidental infections caused in avian
influenza laboratories are in the human workers. For the coronaviruses, even though additional respiratory
protection is worn under BSL3 conditions, most infections are caused by aerosol exposure because other
routes are unlikely to cause an infection. Although this analysis produces a robust estimate of relative risk
in a variety of informative ways, the data used are insufficient to predict absolute risk. A separate method
is used to support a rough estimate of absolute risk in Sec. 6.8.
In sum, the analysis of these release pathways enables the estimation of the relative risk of working with
the GoF pathogens and how the change of any phenotype would alter this risk. Table 6.2 shows the
relative probability (compared to work with seasonal influenza) of a laboratory acquired infection (that
produces some hazard of a causing a local outbreak) when working with the various pathogens considered
in this study. Our analysis considers that vaccination of laboratory workers could reduce the chance of a
laboratory infection and that antivirals could be given prophylactically if a high risk exposure event

occurs. Moreover, health monitoring and isolation protocols would greatly reduce the chance that a
worker mingles with the general population, causing secondary cases and sparking an outbreak. In this
section, these factors are always considered when examining the pathways that lead to a laboratory
acquired infection, because if the infected worker poses no hazard to the population, the consequences of
the accident end with that person.
Table 6.2. Relative probability of a laboratory acquired infection for the various pathogens considered in
this study as compared to work with seasonal influenza.
Pathogen Biosafety Level Relative Probability of an LAI*
Seasonal influenza virus BSL2 1 (defined)
Pandemic influenza virus BSL3 0.10 (0.07-0.15)
Avian influenza virus BSL3 0.43 (0.21-0.90) (mostly of birds)
SARS-CoV BSL3 0.03 (0.02-0.04)
MERS-CoV BSL3 0.01 (0.006-0.02)
These data are generated by comparing the sums of the frequency of infection from all loss of containment
pathways for each pathogen. In this case, we use the term laboratory acquired infection to include an infection of
wild birds to capture the comparative risk of working with avian influenza viruses. The numbers in the
parenthesis are the results from the p5 and p95 outputs of the Monte Carlo analysis.
The irreducible uncertainty in the pathways that lead from laboratory incidents to infections of wild birds
with avian influenza is evident in these results. As will be described below, if infected material leaves the
laboratory, it is assumed that wild birds will access it at the dump because there is no way to estimate
what percent of bags are accessed by birds in a dump. The estimate here is therefore conservative but,
even with the uncertainty provided, suggests that the probability of a wild bird becoming infected in a
laboratory accident with avian influenza is roughly equivalent (within an order of magnitude) to the
probability a person will be infected by a laboratory accident involving seasonal influenza. In contrast, the
risk of an accident leading to an infection with any other pathogen is roughly one (for pandemic
influenza) or two orders of magnitude less (for the coronaviruses).
The sections that follow explore the various pathways that lead from a laboratory accident to a laboratory
acquired infection for each of the pathogens examined. The relative probability of laboratory acquired
infections when working with pathogens with GoF phenotypes compared to the work with wild type
pathogens is described.
6.4.4.1 Laboratory Acquired Infections and Seasonal Influenza
When working with seasonal influenza under BSL2 conditions, the accidental generation of aerosols
produces the majority of laboratory acquired infections because no personal respiratory protection is worn
(and the agent is extremely infectious). Only a small minority of accidental infections are caused by the
contamination of the hands. Figure 6.14 shows the various accident pathways that contribute to the
probability of a laboratory acquired infection. Data in these figures comes from comparing the total
frequencies of laboratory acquired infections, which in turn is derived from the predicted frequency of
exposure events with various pathogen amounts as calculated by the Fault Tree Models. Comparing the
5th, 50th and 95th percentile358 of our Monte Carlo simulations suggests that changes in risk of less than a
factor of two are not that significant because the total probability of an infection from any cause changes
358 Recall that the p50 is the median result, whereas the p95 is the result in which 95% of all results have a smaller value, and
the p5 is the result in which 5% of all results have a smaller value.

by this much between the samples (relative to the p50, the p5 is 2.3-fold less and the p95 is 1.6-fold
greater). In terms of the incidents that contribute to the probability of infection, the 5th and 95th percentile
results are similar to the p50 but the fomite-based pathways contribute to the infections slightly more
frequently (splashes cause from 0.5 to 3% of infections and cuts cause from 0.8% to 10% of infections
depending on the sample).
Figure 6.14. A pie chart showing how the various accident pathways contribute to the total probability of a
laboratory acquired infection for seasonal influenza. Solid colored sections are fomite-based hazards, hatched
sections are aerosol-based hazards and stippled sections are both fomite- and aerosol-based hazards. The
median result of the Monte Carlo analysis is shown.
Some of the GoF phenotypes could affect risk of a laboratory infection (Table 6.3). Viruses that grow to a
high titer could increase the dose received by a victim via an aerosol or fomite-based exposure and
approximately double the chance of a laboratory acquired infection if cultures with these high titers are
routinely manipulated. That being said, many strains of seasonal influenza already grow to a titer of
1E8/ml and increasing this titer may not be desirable or scientifically achievable. The strain could be
made antiviral resistant, which would vitiate providing antivirals after a high-risk exposure. Similarly, the
strain could be made to evade the protection afforded by vaccines. Because seasonal influenza is already
adapted to humans, this GoF phenotype is not relevant for this pathogen. Table 6.3 shows the relative
increase in the probability of a laboratory acquired infection predicted if modified strains of seasonal
influenza are created. As titer increases, splashes begin to contribute more to the risk of a laboratory
acquired infection, but still contribute less than 20% of the total risk (not shown).

Table 6.3 Increase in the probability of a laboratory acquired infection
associated with GoF phenotypes in seasonal influenza.
Phenotype Increase in Probability of a LAI
Evasion of vaccines +50%
Antiviral resistance +40%
Growth to 1E9/ml +100%
Adaptation to humans N/A
6.4.4.2 Laboratory Acquired Infections and Pandemic Influenza
When working with pandemic influenza under BSL3 conditions, the accidental generation of aerosols
produces the majority of laboratory acquired infections even though personal respiratory protection is
worn. About 20% of accidental infections are caused by the contamination of the hands. This finding
holds across the p5 and p95 samples (although in the p95 the fomite-pathways contribute to ~30% of risk)
Figure 6.15 shows the various accident pathways that contribute to the probability of a laboratory
acquired infection.
laboratory acquired infection for pandemic influenza. Solid colored sections are fomite-based hazards,
hatched sections are aerosol-based hazards and stippled sections are both fomite- and aerosol-based hazards.
The median result of the Monte Carlo analysis is shown.
Some of the GoF phenotypes could affect risk of a laboratory infection. Viruses that grow to a high titer
could increase the dose received by a victim via an aerosol or fomite-based exposure. The strain could be
strain could be made to evade the protection afforded by vaccines. Because pandemic influenza is already
adapted to humans, this GoF phenotype is not relevant for this pathogen. Table 6.4 shows the relative
increase in the probability of a laboratory acquired infection predicted if modified strains of pandemic
influenza are created. Enhancing the growth of pandemic strains to achieve titers of 1E9 or 1E10/ml can
significantly increase the risk that a laboratory acquired infection would occur because the exposures that
drive risk are normally very low. That being said, some strains of pandemic influenza already grow to a

titer of 1E8/ml and increasing this titer may not be desirable or scientifically achievable. Increasing the
maximum titer of poor growing strains to 1E8/ml simply allows these strains to approach the risk
modeled for the more robust strains.

associated with GoF phenotypes in pandemic influenza.
6.4.4.3 Laboratory Acquired Infections and Avian Influenza
When working with avian influenza under BSL3 conditions, the accidental release of improperly
decontaminated solid waste drives the risk of an accidental infection, albeit of a wild bird, not a human. In
these cases, the operator committed an error, such as packing the autoclave too tightly with bedding-
containing cages or carcasses such that the steam did not penetrate into all parts of the waste.
Alternatively, the operator could run an improper cycle such that the temperature was not reached for the
required length of time. The waste then enters the solid waste stream and is dumped, whereupon wild
birds (like gulls that frequent garbage dumps) access the infectious material and are infected. Data is
lacking to determine the percent of waste containers actually accessed by gulls, or even how an outbreak
would unfold if gulls that live in garbage dumps were infected; however, the analysis assumes that an
avian outbreak would occur with attendant human infections and deaths from exposure to infected wild or
domestic birds. Direct infections of workers in the laboratory represent less than 25% of the probability of
an infection. Figure 6.16 shows the various accident pathways that contribute to the probability of a
laboratory acquired infection.

laboratory acquired infection for avian influenza (including infections of wild birds). Both solid waste
pathways infect wild birds only, and not humans. Solid colored sections are fomite-based hazards, hatched
sections are aerosol-based hazards and stippled sections are both fomite- and aerosol-based hazards. The
median result of the Monte Carlo analysis is shown.
The differences between the p50 and p5 result illustrate some of the significant uncertainty of the causes
of accidents when working with avian influenza viruses (Figure 6.17). Although the pathways that lead to
infection of a laboratory worker (compared to a bird) begin to contribute more to the probability of
accidents, these pathways still contribute to a minority of infections.
laboratory acquired infection for avian influenza (including infections of wild birds) for the p5 result of the
Monte Carlo analysis. Both solid waste pathways infect wild birds only, and not humans. Solid colored
sections are fomite-based hazards, hatched sections are aerosol-based hazards and stippled sections are both
fomite- and aerosol-based hazards.

Some of the GoF phenotypes could affect risk of a laboratory infection. Viruses that grow to a high titer
could increase the dose received by a victim via an aerosol or fomite-based exposure. The strain could be
strain could be made to evade the protection afforded by vaccines. If the strain were adapted to humans,
we assume it would poorly infect birds but would greatly decrease the infectious dose in humans. Table
6.5 shows the relative increase in the probability of a laboratory acquired infection predicted if modified
strains of pandemic influenza are created. No GoF phenotype increases the risk that an accidental
infection occurs with the avian influenza viruses, because so much of the risk of an accidental infection
wild type pathogen is driven by the infection of birds from solid waste (all the GoF phenotypes affect the
human health risk). In fact, adapting the strain to humans DECREASES the probability that a laboratory
accident will lead to an infection of an animal or person by 30% because although the strain is more likely
to infect a person, it is much less likely to lead to a dangerous outbreak in birds, which can sicken much
more than a handful of laboratory workers. Note, because this analysis considers one GoF trait at a time,
the adaptation to humans is assumed to create a strain that is more infectious in humans but not alter its
transmissibility.

associated with GoF phenotypes in avian influenza.
Adaptation to humans -30%
Figure 6.18 shows the accident pathways that lead to human infections for avian influenza strains adapted
to infect humans (instead of birds).

laboratory acquired infection for avian influenza adapted to infect humans. Solid colored sections are fomite-
based hazards, hatched sections are aerosol-based hazards and stippled sections are both fomite- and aerosol-
based hazards. The median result of the Monte Carlo analysis is shown.
6.4.4.4 Laboratory Acquired Infections and Coronaviruses
When working with the coronaviruses under BSL3 conditions, the accidental generation of aerosols
produces the vast majority of laboratory acquired infections even though personal respiratory protection is
worn. Working with infected animals poses minimal risk because mouse adapted strains poorly infect
human cells due to changes in the spike protein. Figure 6.19 shows the various accident pathways that
contribute to the probability of a laboratory acquired infection, the p5 and p95 results from the Monte
Carlo analysis are similar (not shown).

laboratory acquired infection for SARS-CoV (the chart for MERS-CoV is very similar). Solid colored
sections are fomite-based hazards, hatched sections are aerosol-based hazards and stippled sections are both
fomite- and aerosol-based hazards. The median result of the Monte Carlo analysis is shown.
Only one of the GoF phenotypes could affect risk of a laboratory infection. Viruses that grow to a high
titer could increase the dose received by a victim via an aerosol or fomite-based exposure. Because the
coronaviruses are already adapted to humans, and because there are no countermeasures in use for
protecting against infections with this pathogen, other GoF phenotypes are not relevant for this pathogen.
Table 6.6 shows the relative increase in the probability of a laboratory acquired infection predicted if
modified strains of the coronaviruses are created. Enhancing the growth of the coronaviruses to achieve
titers of 1E9 or 1E10/ml can significantly increase the risk that a laboratory acquired infection would
occur because the exposures that drive risk are normally very low. Under these circumstances,
contamination of the hands beings to significantly drive risk, growing to cause about 20% of all
laboratory infections for strains that grow to 1E10/ml (not shown). That being said, SARS- and MERS-
CoV already grow to a titer of 1E8/ml and increasing this titer may not be desirable or scientifically
achievable.
Table 6.6 Increase in the probability of a laboratory acquired infection associated

with GoF phenotypes in the coronaviruses. N/A marks a phenotype not applicable
to the coronaviruses.
Evasion of vaccines N/A
Growth to 1E9/ml +260% (SARS-CoV), +160% MERS-CoV
Growth to 1E10/ml +860 (SARS-CoV), +550% (MERS-CoV)

6.4.4.5 Effect of various research conditions on risk on probability of loss of containment
6.4.4.5.1 Effect of changing the biosafety level when working with GoF pathogens
This section describes how changing the biosafety level of the laboratory in which GoF pathogens are
manipulated changes risk. All GoF pathogens except for seasonal influenza are manipulated at BSL3
containment at least. Increasing the containment level of seasonal influenza decreases the probability of a
laboratory acquired infection by three-fold, which, notably, would partially compensate for the increases
in risk caused by the riskiest GoF phenotypes. For all but avian influenza, the increase or decrease in risk
is caused by the addition or elimination of personal respiratory protection (such as PAPRs). For avian
influenza, fewer mistakes need to be committed to release infected solid waste at BSL2 than BSL3
leading to an increase in the frequency of infections of wild birds.
Table 6.7. Change in the probability of a laboratory acquired infection (LAI) for changes in the
containment level required for manipulating the GoF pathogens.
Pathogen Change in BSL Change in Probability of an LAI
Seasonal influenza Increase from BSL2 to BSL3 3-fold decrease
Pandemic influenza Decrease from BSL3 to BSL2 3.5-fold increase
Avian influenza Decrease from BSL3 to BSL2 110-fold increase
SARS-CoV Decrease from BSL3 to BSL2 Less than 2-fold increase
MERS-CoV Decrease from BSL3 to BSL2 Less than 2-fold increase
In contrast, if any of the other GoF pathogens were manipulated under BSL2 conditions instead of BSL3,
the probability of a laboratory acquired infection would, unsurprisingly, increase, although this increase is
small for the coronaviruses. This analysis suggests that work on influenza viruses in parts of the world
with less stringent biosafety standards than the US could be expected to have up to an order of magnitude
more accidents resulting in an infection.
6.4.4.5.2 Factors that influence the probability of accidents with seasonal influenza virus
To understand which laboratory features and practices influence the probability of a laboratory acquired
infection with the risk of causing an outbreak, a sensitivity analysis was performed in which the values of
any parameter were set to the lowest or highest level while all other parameter values were allowed to
vary as normal. The results of this sensitivity analysis for seasonal influenza at BSL2 are shown in the
one-sided tornado plot in Figure 6.20 wherein the width of the boxes shows the increase in the probability
of an infection if the parameter is set from its value that minimizes the probability to the value that
maximizes it.

Figure 6.20. A one-sided tornado plot that shows the increase in the probability that a laboratory accident
with seasonal influenza virus would lead to an infected individual mingling with the community. The left side
of each box (set to one) represents that value of that parameter that minimizes this probability, whereas the
right side is the value that maximizes this probability.
The most influential features that influence the risk of an infection occurring and that worker posing a risk
to the community is the behavior of that worker. Maximizing the probability that a worker will not
properly report a high-risk exposure can increase the probability of a dangerous infection by a three-fold.
Similarly, maximizing the chance that a worker violates isolation protocols after an overt exposure can
increase risk by seven-fold. For this reason, extensive training on the benefit of reporting, health
monitoring and isolation could increase compliance and greatly reduce risk. No other parameter is very
influential (partially because respirators are not worn in BSL2).
6.4.4.5.3 Factors that influence the probability of accidents with pandemic influenza virus
The results of the sensitivity analysis for pandemic influenza at BSL3 are shown in the one-sided tornado
plot in Figure 6.21 wherein the width of the boxes shows the increase in the probability of an infection if
the parameter is set from its value that minimizes the probability to the value that maximizes it.

with pandemic influenza virus would lead to an infected individual mingling with the community. The left
side of each box (set to one) represents that value of that parameter that minimizes this probability, whereas
the right side is the value that maximizes this probability.
The most important practice of reducing the probability of a dangerous infection with pandemic influenza
is the isolation of possibly infected workers (poor isolation practices increase the risk of an infection by
four-fold). Similarly, poor reporting of exposure can more than double the probability of a double
infection. Failure to double glove can more than double the probability of an infection, whereas poorly
functioning or fitted respirators can nearly double this probability.
6.4.4.5.4 Factors that influence the probability of accidents with avian influenza virus
The results of the sensitivity analysis for avian influenza at BSL3 are shown in the one-sided tornado plot
in Figure 6.22 wherein the width of the boxes shows the increase in the probability of an infection if the
parameter is set from its value that minimizes the probability to the value that maximizes it. No feature or
practice in the assessment conducted influences the probability of a dangerous infection by more than 1.5-
fold, which makes sense because most of the risk is driven by errors in solid waste processing.

with avian influenza virus would lead to an infected individual mingling with the community. The left side of
each box (set to one) represents that value of that parameter that minimizes this probability, whereas the
right side is the value that maximizes this probability.
6.4.4.5.5 Factors that influence the probability of accidents with coronaviruses
The results of the sensitivity analysis for the coronaviruses at BSL3 are shown in the one-sided tornado
plot in Figure 6.23 wherein the width of the boxes shows the increase in the probability of an infection if
the parameter is set from its value that minimizes the probability to the value that maximizes it.

with SARS-CoV would lead to an infected individual mingling with the community. The left side of each box
(set to one) represents that value of that parameter that minimizes this probability, whereas the right side is
the value that maximizes this probability.
Three practices have a significant influence on the probability that an infection occurs and the worker
mingles with the community. Firstly, failure to double glove can increase the probability by up to four-
fold. From the same exposure pathways, poor hand washing can increase the probability by nearly 50-
fold. These findings demonstrate that worker education and training on proper techniques for reducing
hand contamination may significantly reduce risk of working with the coronaviruses. Also, poor
adherence to isolation protocols can increase the probability that an infected worker mingles with the
population by ten-fold. Once again, training on the importance of health monitoring and isolation could
greatly reduce risk.
The results for MERS-CoV follow the same overall trends, and are shown in Figure 6.24, below.

with MERS-CoV would lead to an infected individual mingling with the community. The left side of each box
(set to one) represents that value of that parameter that minimizes this probability, whereas the right side is
the value that maximizes this probability.
6.4.4.5.6 Importance of Laboratory Worker Training
By far, the most critical driver of the probability of a dangerous infection is the behavior of workers
themselves. As discussed above, the probability that a worker reports a high risk exposure or adheres to
isolation protocols can significantly influence the probability that an infected worker would mingle with
the population. However, the probability that a worker would carelessly or forgetfully cause the incident
in the laboratory is the most influential factor on risk. Figure 6.25 shows the relative influence of
parameters influencing human error rates in the laboratory against all other parameters investigated. In
this instance, all nodes in the fault trees that were based on the probability of a human error occurring had
their failure probabilities (i.e., the probability a mistake is committed) simultaneously set to their
maximum or minimum values. Only human errors that occur within the laboratory, leading to an accident,
were considered; the probabilities of human errors occurring after an incident occurs, such as failures to
report incidents or remain in isolation, were unchanged in this analysis. From this figure, human error
rates can influence the probability of an infection by more than 100-fold (whereas the next most
influential parameter for seasonal influenza changes this probability by nearly 10-fold). Across all
pathogens studied, human error rates in the laboratory this type of parameter influence the probability of a
dangerous infection from 100-1,000-fold (data not shown).

with seasonal influenza would lead to an infected individual mingling with the community. The left side of
each box (set to one) represents that value of that parameter that minimizes this probability, whereas the
right side is the value that maximizes this probability. The x-axis is on a log scale.
This analysis reflects both aleatory and epistemic uncertainty. That is, data are lacking on how often
humans will make mistakes of a variety of kinds in a biological laboratory (epistemic uncertainty).
However, even if relevant observational data were available to inform these human error rates, significant
aleatory uncertainty would remain. That is, laboratory workers are humans and some humans are more
prone than others to errors due to carelessness, unfamiliarity with protocols, distraction or stress. Many
who have experience working in a microbiology laboratory could identify co-workers with whom no one
would share reagents due to the perception that the co-worker would contaminate or otherwise
compromise the reagent. Aleatory uncertainty will always exist because at any given time, it is unknown
which type of person will be working in the laboratory (and what stresses they will be under). This
analysis suggests that efforts to reduce stressors on workers could significantly improve laboratory safety.
Also, measures to identify and re-train workers that are prone to carelessness or forgetfulness may have
similar benefits. Moreover, as described by the laboratory safety stakeholders we interviewed, efforts to
“train-in” to a BSL3 laboratory by first demonstrating competence and mastery of protocols in a BSL2
laboratory could significantly improve safety. Lastly, some stakeholders mentioned that dedicated
professionals handle some sensitive laboratory protocols, such as the operation of autoclaves, to reduce
the probability of the release of contaminated materials. Such practices would also significantly improve
safety.
6.5 Consequences of an Outbreak Caused by an Avian Influenza Strain that Is Not

Transmissible in Mammals
As discussed in the methods, we are unable to adequately model the human health consequences of an
outbreak of an influenza virus that is not transmissible amongst people but is maintained in birds. Our
simple models, based on the characteristics of past avian influenza outbreaks, suggest that an average of

100 people would die and 1,000 people would be clinically ill from contact with infected wild birds or
poultry.
Most GoF phenotypes would not affect risk (clearly, if the strain were made transmissible in mammals,
risk could change greatly as explored in Sec. 6.7 below). Enhanced growth in culture would not affect the
outbreak unless this trait was related to pathogenicity or infectiousness. Ability to overcome immunity
would not increase risk because most humans have no prior immunity from exposure to avian strains and
novel vaccines are not stockpiled in quantity for an outbreak of influenza that is not human transmissible.
Resistance to antivirals is of minimal risk because some wild type strains of avian influenza are already
resistant and antivirals at most would reduce the number of deaths by half (and the role of antivirals in
preventing onward transmission is moot).
No prediction was able to be made on how adaptation to a mammalian host, which could reduce the
median infectious dose, affects risk. The infectious dose of any given strain of avian influenza in humans
is unknown as is the dose to which past victims had been exposed to. It is possible that upon exposure to
an infected bird, a human receives either a large dose or no dose at all (if, for example, infection is
generally caused by inoculation with a globule of infected feces that could contain billions of active virus
particles). In this case, a reduction of the median infectious dose would minimally affect risk. Conversely,
many people could be exposed to very low doses and not become infected; if the infectious dose
decreased perhaps all these people would develop illness.
In a simple way, the effect of an increased case fatality rate on risk can be made. Figure 6.26 shows the
human deaths predicted to occur from an outbreak of wild type avian influenza, an outbreak caused by a
strain that causes an illness with half the rate of survival of wild type avian influenza, and an outbreak that
causes an illness caused by a strain with quarter the rate of survival of wild type avian influenza.
Decreases in rates of survival must be modeled instead of increases in death rates because the fatality rate
could be as high as 50% in some wild type strains. Outbreaks of wild type avian influenza are predicted to
cause about 100 deaths, and very few outbreaks would cause up to 500 deaths. If a strain were modified
to decrease the survival rate in victims by half, the outbreaks cause about 300 deaths on average, but up to
700 deaths in rare cases. If a strain were modified to decrease the survival rate in victims by a quarter, the
outbreaks cause about 500 deaths on average, but up to 900 deaths in rare cases. This analysis presumes
that the most pathogenic strains of avian influenza have an inflated case fatality rate due to the under-
reporting of mild cases. If the case fatality rate of the most pathogenic strains of avian influenza is truly
50% then increasing this trait would not make the strain much more dangerous.

Figure 6.26. The number of outbreak simulations (out of 1,000 per condition) resulting in a number of deaths
for wild type avian influenza and strains modified to be more pathogenic. Outbreaks of wild type avian
influenza are in blue, outbreaks of a strain that causes disease with half the survival rate of wild type avian
influenza are in orange, and outbreaks of a strain that causes disease with quarter the survival rate of wild
type avian influenza are in grey.
That being said, the state of modeling of avian influenza outbreaks in human populations is very
rudimentary so confident predictions of the risk of modified avian influenza strains in human populations
is currently impossible. Additional data on the risk factors that lead to human infection, the life cycle of
disease in its various avian hosts and factors that relate the biology of the virus to the pathology in the
various hosts is needed to improve modeling of this infectious disease.
6.6 Risk of an Outbreak Escaping Local Control of Pathogens that Are Transmissible in
Mammals
After an outbreak is initiated, the GoF phenotype of enhanced growth characteristics in culture or eggs no
longer has any influence on risk. If this phenotype also increases transmissibility in humans, then the
models capture changes in risk through those parameters. Pathogenicity is indirectly captured in the
probability that an outbreak escapes local control. That is, the strength of public health control measures
and social distancing exerts a critical influence on the probability that an outbreak escapes local control,
which is assumed to be stronger when the outbreak is causing significant mortality than when the
outbreak resembles a typical influenza season. Pathogenicity directly influences consequences in terms of
the number of deaths that occur should an outbreak not escape local control (if the outbreak escapes, then
the consequences will be dominated by the global deaths summed across all regions, not the deaths in one
community). Ability to overcome immunity induced by vaccination is not relevant because a matched
vaccine will not be available in quantity in time to respond to the initial outbreak.
Antivirals have never been dispensed to address a nascent outbreak of influenza and public health
authorities interviewed have no concrete plans for the use of antivirals in an outbreak arising from a local
laboratory. For this reason, we do not know if, in the case of a laboratory-associated outbreak, antivirals
would be mass dispensed to the entire outbreak area, if they would be distributed to all contacts of an
infected person or if they would just be given to the infected individuals. Moreover, some strains of
influenza are naturally resistant to antivirals and we do not know which strain would be involved in an
outbreak. For these reasons, antivirals were not included in the branching process model. Because data
exists on how antivirals are used in the context of an ongoing global pandemic of influenza, antivirals are
included in the global influenza models described in Sec. 6.11, below. Similarly, antivirals can be given

upon high-risk exposures in the laboratory to prevent the onset of illness or reduce transmissibility if an
infection occurs, as described in Sec. 6.4.
All of the figures shown assume that just one person is initially infected. These events dominate risk
because they are much more likely to occur and have similar consequences to events that initially infect
multiple people. As discussed in Figure 6.27 below, increasing the number of initially infected at most
increases the probability of a global pandemic by ten-fold. However, events that lead to a single initial
infection are more than 100-fold more likely to occur.
Even if an infected person mingles with the local population, secondary infections in the population are
not guaranteed. In fact, for some poorly transmissible pathogens (or the coronaviruses that have a high
variance in transmissibility), in most cases no secondary cases are caused just by chance. Figure 6.27
shows the relationship between transmissibility and the percent of outbreaks that create at least one
secondary infection for the influenza viruses and the coronaviruses. When a single person infected with
seasonal influenza mingles with the population, another person is infected just half the time (and this
probability increases modestly as R0 increases). In contrast, when a single person infected with a SARS-
like disease mingles with the population, at least one secondary case is caused only 30% of the time,
which is expected given the high variance of the transmissibility of that disease. If two infected people
mingle with the population, the chances that at least one secondary infection is caused increases. Perhaps
most importantly, as transmissibility increases dramatically, the probability of at least one secondary
infection increases only modestly, by less 15% for an increase in R0 of one (except for very low values of
R0). The chance that an infected person does not cause any secondary infections is integrated into the
analysis of outbreaks escaping local control described below.
Figure 6.27. The probability that at least one secondary infection is caused by one (left panel) or two (right
panel) infected people mingling with the population for various wild type and enhanced viruses. The R 0s used
in this figure are 0.1-0.2 for wild type avian influenza, 0.2-1.0 for enhanced avian influenza, 1.0-1.2 for
enhanced avian influenza+, 1.2-1.4 for wild type seasonal influenza, 1.4-1.9 for wild type pandemic influenza,
1.9-2.2 for enhanced pandemic influenza, 2.2-2.5 for enhanced pandemic influenza+, 0.4-0.6 for wild type
MERS-CoV, 0.8-1.2 for wild type SARS-CoV low, 1.2-1.6 for wild type SARS-CoV high, 1.6-1.9 enhanced
SARS-CoV, 1.9-2.2 enhanced SARS-CoV+, 2.2-2.5 for enhanced SARS-CoV++. “Error bars” show the range
of results across 80% of the parameter values explored in the analysis (the parameter values that cause the
highest and lowest 10% of results are not represented).

6.6.1 Effect of enhanced transmissibility in mammals on risk of an outbreak escaping local control
Transmissibility has a significant influence on the chance that an outbreak would escape local control for
all GoF pathogens.
6.6.1.1 Seasonal influenza
If a single person is initially infected by a loss of containment event with a seasonal influenza strain that
has not circulated recently and the infected person mingles with the general population, stochastic forces
and control measures still cause the outbreak to extinguish the vast majority of the time. The tornado plot
in Figure 6.28 illustrates how a variety of parameters influence the probability that an outbreak would
escape local containment (the wider the box, the more influence that parameter value has on the outbreak
escaping local control). This figure shows that an outbreak caused by a single infection with wild type
seasonal influenza virus has a 20% chance of escaping local control. Of wild type strains, those with the
highest R0 values (1.4) have up to a 30% chance of escaping local control. If transmissibility were
increased even further (to 2.2), the probability of an outbreak escaping local control could more than
double to 60%.
Figure 6.28. A tornado plot showing how values of modeling parameters affect the probability that an
outbreak of seasonal influenza would escape local control if a single person were initially infected (shown on
the Y-axis). Open boxes represent the range of probabilities that an outbreak would escape local control for
all parameter values sampled for the wild-type pathogen whereas grey boxes represent possible
enhancements in a GoF strain. The vertical line shows the median result across all parameter values for an
outbreak caused by a wild type strain.
Although the probability of an outbreak escaping local control is sensitive to the transmissibility of the
strain that causes the outbreak, this increase poses a risk only if the creation of a strain with such
properties is feasible. Figure 6.29 shows the relationship between the transmissibility of a seasonal
influenza strain and the probability that a resulting outbreak would escape local control. These data show
that modifying a wild type seasonal influenza strain associated with an average R0 value (1.3) so that it
has the transmissibility associated with an average pandemic influenza strain (1.7) doubles the probability
that an outbreak would escape local control. Also of note, the least transmissible wild type strains (R0 of
1.2) are roughly two-to-three-fold less likely to cause an outbreak that escapes local control as the most
transmissible wild type strains (R0 of 1.4).

Figure 6.29. The relationship between transmissibility of seasonal influenza virus (as measured by the R 0 of
the resulting outbreak) and the probability that an outbreak escapes local control. Grey represents various
manipulations to increase the transmissibility of the virus beyond estimates for wild type strains. The colors
correspond to those represented in the tornado plot in Figure 6.28. “Error bars” show the range of results
across 80% of the parameter values explored in the analysis (the parameter values that cause the highest and
lowest 10% of results are not represented). The error bars reflect the effect of an uncertain k value (the
variance in transmissibility of the disease among those infected) on the probability that an outbreak escapes
local control. For this reason, for any particular k value, the relative increase in the probability that an
outbreak escapes local should mirror the shape of the line given for the median result.
6.6.1.2 Pandemic influenza
If a single person is initially infected by a loss of containment event with a pandemic influenza strain and
that infected person mingles with the general population, stochastic forces and control measures still
cause the outbreak to extinguish the majority of the time. Figure 6.30 shows that an outbreak caused by a
single infection with wild type seasonal influenza virus has a 30% chance of escaping local control. Of
wild type strains, those with the highest R0 values (1.9) have up to a 42% chance of escaping local
control. If transmissibility were increased even further (to 2.5), the probability of an outbreak escaping
local control could double to 60%.

outbreak of pandemic influenza would escape local control if a single person were initially infected. Open
boxes represent the range of probabilities that an outbreak would escape local control for all parameter
values sampled for the wild-type pathogen whereas grey boxes represent possible enhancements in a GoF
strain. The vertical line shows the median result across all parameter values for an outbreak caused by a wild
type strain.
Although the probability of an outbreak escaping local control is sensitive to the transmissibility of the
strain that causes the outbreak, this increase poses a risk only if the creation of a strain with such
properties is feasible. Figure 6.31 shows the relationship between the transmissibility of a pandemic
influenza strain and the probability that a resulting outbreak would escape local control. These data show
that modifying a wild type pandemic influenza strain associated with a median R0 value (1.7) so that it
has a transmissibility greater than any estimate for any influenza strain (2.4) merely increases the
probability that outbreak escapes control by 50%. If however, the local population can sustain robust
social distancing throughout the nascent outbreak (for example, by reducing the number of human
contacts they have by half, shown as a community mitigation of 0.5 in the figure below), these extreme R0
values would be required for the outbreak to have any chance of escaping local control. This finding is
intuitively obvious because halving all contacts would reduce an R0 value of two to an R0 of one, which is
required for the outbreak to be self-sustaining.

Figure 6.31. The relationship between transmissibility of a pandemic influenza virus (as measured by the R0
of the resulting outbreak) and the probability that an outbreak escapes local control. Grey indicates various
manipulations to increase the transmissibility of the virus beyond estimates for wild type strains. The colors
correspond to those represented in the tornado plot in Figure 6.30. “Error bars” show the range of results
lowest 10% of results are not represented). The error bars reflect the effect of an uncertain k value (the
variance in transmissibility of the disease among those infected) on the probability that an outbreak escapes
local control. For this reason, for any particular k value, the relative increase in the probability that an
outbreak escapes local should mirror the shape of the line given for the median result.
6.6.1.3 Avian influenza
Wild type avian influenza virus is insufficiently transmissible in mammals to cause an outbreak that
escapes local control. Figure 6.32 shows how transmissible in people a modified strain of avian influenza
would have to be to escape local control should one laboratorian be initially infected and mingle with the
general population. Unless robust social distancing measures can be implemented throughout the outbreak
(community mitigation 0.5 in the figure below), increasing the transmissibility of an avian influenza strain
in humans to that of seasonal influenza would lead to a local outbreak with about a 10-20% chance of
escaping local control. Given that the wild type strain has no chance of creating an outbreak that escapes
local control (or even one that is made modestly more transmissible) this increase is extremely significant.

Figure 6.32. The relationship between transmissibility in humans of an avian influenza virus (as measured by
the R0 of the resulting outbreak) and the probability that an outbreak escapes local control. Grey points
indicate various manipulations to increase the transmissibility of the virus beyond estimates for wild type
strains. “Error bars” show the range of results across 80% of the parameter values explored in the analysis
(the parameter values that cause the highest and lowest 10% of results are not represented). The error bars
reflect the effect of an uncertain k value (the variance in transmissibility of the disease among those infected)
on the probability that an outbreak escapes local control. For this reason, for any particular k value, the
relative increase in the probability that an outbreak escapes local should mirror the shape of the line given
for the median result.
6.6.1.4 Coronaviruses
If a single person is initially infected by a loss of containment event with SARS-CoV and that person
mingles with the general population, stochastic forces and control measures still cause the outbreak to
extinguish. Figure 6.33 shows that an outbreak caused by a single infection with wild type SARS-CoV
has nearly no chance of escaping local control. The historical outbreaks of coronaviruses reinforce this
finding because although these outbreaks lead to infections in several locations, they did not initiate a
global pandemic because local control of the outbreak was successful in every outbreak location. The
highest estimates for the value of R0 (1.6) for outbreaks caused by wild type SARS-CoV, cause outbreaks
that have up to a 10% chance of escaping local control. If transmissibility were increased even further (to
an R0 of 2.4), the probability of an outbreak escaping local control could increase significantly to 25%.

outbreak of SARS would escape local control if a single person were initially infected. Open boxes represent
the range of probabilities that an outbreak would escape local control for all parameter values sampled for
the wild-type pathogen whereas grey boxes represent possible enhancements in a GoF strain. The vertical line
shows the median result across all parameter values for an outbreak caused by a wild type strain.
Wild type MERS-CoV is not transmissible enough to cause an outbreak to escape local control. Should a
MERS-CoV be modified to be as transmissible as SARS-CoV, then its probability of escaping local
control would be similar.
As transmissibility of SARS-CoV increases, the probability that an outbreak escapes local control
increases. The relationship between transmissibility and probability of an outbreak escaping is shown in
Figure 6.34. As the R0 of an outbreak approaches the high end of estimates for outbreaks caused by wild
type SARS-CoV, the probability of escape approaches 20% for most values of community control. If
someone were to modify the SARS-CoV to be even more transmissible in humans than these high-end
estimates would indicate, the probability that the disease escapes local control increases, but not
dramatically (this part of the line is shaded in rainbow colors). These data show that even dramatic
increases in transmissibility do not relate to dramatic increases in the chance that the outbreak will
extinguish. However, since there is a minimal risk of wild type SARS-CoV causing a global outbreak,
this increase to a non-zero probability can be significant. The relationship for MERS-CoV is similar to
that shown in Figure 6.34 except that GoF experiments that increase transmissibility must be conducted
for the pathogen to have any chance of creating an outbreak that escapes local control.

Figure 6.34. The relationship between transmissibility of SARS-CoV (as measured by the R0 of the resulting
outbreak) and the probability that an outbreak escapes local control. Grey points indicate various
manipulations to increase the transmissibility of SARS-CoV in humans beyond the highest estimates for the
virus (R0=1.6). “Error bars” show the range of results across 80% of the parameter values explored in the
analysis (the parameter values that cause the highest and lowest 10% of results are not represented).
6.6.2 Effect of enhanced pathogenicity on risk of an outbreak escaping local control
6.6.2.1 Pandemic/seasonal influenza
All influenza outbreaks that extinguish do so when the outbreak is relatively small. For this reason, even
the most highly pathogenic strains influenza would lead to only a handful of deaths (we predict that even
a 1918-like strain would not even one person if the outbreak extinguished locally). If the outbreak
escaped local control and spread throughout the world, vastly more deaths could occur, but these
consequences are assessed in Sec. 6.11.
Interestingly, an outbreak associated with significant mortality may trigger more robust and prolonged
social distancing, which would greatly decrease the chance that an outbreak would spread beyond local
control. For this reason, an outbreak caused by a strain that is modified to be more deadly may actually
reduce risk, although we cannot quantify how the public will react to a novel outbreak.
Wild type avian influenza strains are already associated with a high case-fatality rate and so increasing
this rate would probably have little influence on the robustness of a public health response. Moreover,
wild type avian influenza strains are insufficiently transmissible in humans to cause an outbreak that
would escape local control.
Infections with wild type SARS- and MERS-CoV is already associated with a relatively high case fatality
rate. Increasing this rate is likely to have little influence on the robustness of social distancing. Also,
because the case fatality rate is already significant, increasing this rate has little influence on the number

of deaths expected. For SARS and MERS outbreaks that start with one person that extinguish, we expect
less than 10 people to die even if the strain were modified to be more pathogenic.
6.6.3 Effect of overcoming/evading natural/residual/innate immunity on the probability of an

outbreak escaping local control
6.6.3.1 Pandemic/seasonal influenza
Innate/residual immunity in a population can significantly affect the kinetics of an outbreak of influenza
because prior exposure to recently circulating strains of influenza afford protection to other serotypes.
The protective value of residual/innate immunity is already accounted for in the effective R0, which is one
reason why the R0 for seasonal influenza is significantly less than that of pandemic influenza. In Figure
6.35, below, shows the relation between reduction in innate or residual immunity and the probability that
an outbreak would escape local control for given R0 values.
Figure 6.35. The effect of the evasion of innate/residual immunity on the probability of a seasonal influenza
(lighter blue) or pandemic influenza (darker blue) outbreak escaping local control. The left hand box in each
panel represents the result with the baseline value of prior immunity. In the left-hand panel, the presumption
is that 40% of the population is protected against infection with a wild type strain seasonal. Under this
condition the effect on the probability of escape if a strain (with the same R0 value) were able to overcome
most of this residual/innate immunity (so that only 10% of the population were immune) or overcome all
immunity is shown. In the center and right-hand panels, the presumption is that 10% of the population has
immunity to the wild type strain. Under this condition, the effect on the probability of escape is show for
strains that are modified to overcome all immunity.

This analysis demonstrates that the evasion of pre-existing immunity can significantly increase the
probability of an outbreak of seasonal influenza escaping local control, by two-to-three-fold, if the
population has a high level of residual immunity (as is likely for seasonal influenza since prior
vaccination or illness provides some protection against new strains). Similar to R0, this parameter
influences the probability of an outbreak escaping by enabling the disease to spread more quickly
(because each contact is more likely to result in an infection). Pre-existing immunity can protect a
significant proportion of the population if the strain released is similar (or identical) to a strain of
influenza that recently circulated, which is one reason why this parameter is highly influential. If the
population exhibited relatively low levels of prior immunity, then evasion of prior immunity has little
influence on consequences (increasing the probability of escape by less than a fifth).
Because very few humans have been previously exposed to avian subtypes of influenza and because wild
type strains are poorly transmissible in people, residual immunity has essentially no bearing on the
probability that an outbreak would escape local control.
Because very few humans have been previously exposed to SARS- or MERS-CoV, residual immunity has
essentially no bearing on the probability that an outbreak would escape local control.
6.6.4 Effect of loss of containment pathways on risk of loss of local control of an outbreak
The nature of the incidents that could lead to a loss of containment event affect the probability of an
outbreak escaping local control in three ways:
1. Incidents can be covert or overt and faster implementation of control measures is possible with
overt incidents,
2. Incidents can initially infect a laboratory worker or member of the public, and
3. Incidents can infect a single person or multiple people
In this section we explore how the loss of containment pathway affects probability of an outbreak.
6.6.4.1 Overt vs covert incidents
Some incidents are easy to recognize by the public health and laboratory safety communities as having a
very high probability of causing infections outside the laboratory. In the biosecurity assessment, discussed
in Chapter 7, the self-announcing events include mass shootings and bombings of the laboratory. In the
biosafety assessment, the only event that poses a risk of loss of containment that falls into this category is
the earthquake. If an earthquake strikes a laboratory such that obvious physical damage occurs that
breaches the containment suites, the response community is likely to adopt measures assuming that the
population is at risk of an infection and potential outbreak. Moreover, the community, fearful of the work
done in the laboratory are likely to significantly change their behaviors. Lastly, many laboratory buildings
that house work on wild type influenza- or coronaviruses also house work on other human pathogens, so
any work done on influenza- and coronaviruses may contribute only a portion of the overall risk of such
an event.

If we assume that control measures can be immediately implemented but these control measures are no
stronger than those implemented in a laboratory-based outbreak caused by other events, the probability
than an outbreak escapes local control is decreased only modestly if at all (Figure 6.36).
Figure 6.36. Reduction in the probability that an outbreak would escape local control for outbreaks caused by
self-announcing events (like earthquakes) vs other events (like splashes). The reduction in probability is small
and drops to zero for self-announcing events that initially infect large numbers of people.
If immediate and strong social distancing measures can be adopted (such that people halve the number of
contacts they normally have) when an obvious breach in the laboratory is recognized, then no outbreak
escapes local control. This result may be intuitively obvious because most outbreaks caused by a wild
type influenza virus have an R0 value less than two and this degree of control would drop the R0 below
one, which is required for the outbreak to be self-sustaining. We have no data to determine how people
would behave after a large earthquake destroys a containment laboratory in the context of the chaos
caused by the larger event. Perhaps a catastrophic earthquake would naturally reduce the contact between
people in the community because school and work will be suspended. Alternatively, perhaps large
number of people gathering in shelters would increase the contact between individuals and make outbreak
control extremely difficult.
Due to the irreducible uncertainty and the minimal effect of the implementation of immediate control
measures (which are assumed to be similar in strength to those implemented after a covert loss of
containment event), the biosafety analysis assumes that an outbreak in the aftermath of an earthquake that
destroys the laboratory has the same chance of coming under local control as any other outbreak.
Similarly, in the biosecurity section, since the self-announcing events strike the laboratory with minimal
consequences elsewhere (like a bombing or mass shooting), we presume that immediate control measures
can be implemented to control the resulting outbreak although these have a minimal additional influence
on the outbreak escaping local control.
6.6.4.2 Initial infections of the public vs. initial infections of laboratory workers
If a worker violates the protocol and mingles with the general population while sick, this person has the
nearly the same probability of causing an outbreak that spreads beyond local control as an infected
member of the public (not shown). The fact that a laboratory worker is trained to report early symptoms

of unusual illness, preemptively self-isolate (and potentially receive prophylactic antivirals) significantly
reduces the probability that a worker will not mingle with the general population, as explained above.
6.6.4.3 Initial numbers of people infected
Depending on the loss of containment pathway, one, two or more people could be infected by the event.
The vast majority of loss of containment events lead to the infection of one laboratorian who
contaminated her hands, failed to decontaminate them thoroughly and then infected herself and no one
else due to the contamination. However, some loss of containment events lead to multiple people infected
either directly (via aerosols generated inside the laboratory) or indirectly (via a contaminated worker who
happens to physically contact several people soon after leaving the laboratory). Figure 6.37 shows how
the probability of an outbreak escaping local control depends on the initial number of people infected for
seasonal influenza, pandemic influenza and SARS. If one person is infected with influenza and mingles
with the local population, the outbreak has a 20-30% chance of seeding a global pandemic. As more
people are initially infected, the outbreak has a much greater chance of growing beyond local control.
Even with 100 initially infected individuals, a SARS outbreak has a minimal chance of escaping local
control (unless the R0 for the pathogen is at the high end of all estimates). In addition, a SARS outbreak
has a minimal chance of seeding a global pandemic due to the efficacy of control measures at preventing
its spread (unless it is the R0 for an outbreak in the US is at the high end of estimates of R0s estimated for
this pathogen).
Figure 6.37. The relationship between the probability of an outbreak expanding beyond local control and the
number of people initially infected by the loss of containment event. In this figure the median probability of
an outbreak not extinguishing across all parameters for seasonal influenza, pandemic influenza and SARS
are shown. The X-axis is on a log scale (the data points are for 1, 2, 5, 10 and 100 initial infections).
As explained above, the probability that a loss of containment event leads to the initial infection of one
person is much more than ten times as likely as an event that initially infected multiple people. Since the
probability of an outbreak escaping local control is not an order of magnitude greater for outbreaks in
which more than one person is initially infected, incidents that infect exactly one person dominate the risk
of a global outbreak. That is, because incidents that create exactly one index infection happen much more
frequently than incidents that create multiple index infections, yet are still relatively likely to cause a
global outbreak, these incidents are responsible for most of the global pandemics modeled.

6.7 Consequences of a Global Pandemic of Pathogens that Are Transmissible in
Mammals
This section provides a description of the effect of GoF experiments on the consequences of a global
pandemic. Because the relative risk of changing any phenotype depends upon the type of pathogen being
modified (and its wild type traits), the phenotypes that have the most influence on risk for each pathogen
are summarized first. In the sections that follow, a description is provided on exactly how risk changes as
those phenotypes are altered.
After an outbreak is initiated, the GoF phenotype of enhanced growth characteristics in culture or eggs no
longer has any influence on risk. If this phenotype also increases the transmissibility or pathogenicity in
humans, then the models capture changes in risk through those parameters. Moreover, regardless of how
the outbreak began, once it has spread globally the consequences of the global pandemic depend on the
characteristics of the pathogen, not the means by which the outbreak was initiated.
6.7.1 Seasonal Influenza Virus
Even if a wild type strain of seasonal influenza sparked a global outbreak the consequences of this
pandemic would eclipse those from all industrial accidents ever suffered. This section describes which
GoF phenotypes would influence the consequences of a global outbreak of seasonal influenza strains. The
GoF phenotypes relevant to an ongoing global outbreak of seasonal influenza are:
• The ability to overcome protective vaccination,
• The ability to overcome prior immunity (either natural or induced by previous vaccinations or
infection by similar strains in the past),
• Resistance to antivirals,
• Transmissibility, and
• Pathogenicity (used here, case fatality rate)
The ability to evade diagnostics is of secondary importance to the effect of antivirals because diagnostics
are used primarily to direct limiting stocks of these antivirals to only those truly infected. Few other
effects of evading diagnostics exist at this stage of the outbreak because the agent causing the outbreak
would already be identified by the time the outbreak has spread globally and mass vaccination (as soon as
a protective vaccine were available) would occur instead of vaccination based on identified cases.
Moreover, public health resources are insufficient for case isolation and quarantine when an outbreak has
become global.
The GoF phenotypes of enhanced growth is irrelevant to an ongoing outbreak unless it influences
transmissibility or pathogenicity (the risk of which are analyzed here). Adaptation to mammalian hosts is
irrelevant for this pathogen because it is already adapted to infect and spread amongst humans.
To understand how various GoF phenotypes influence the consequence of a global outbreak, a sensitivity
analysis was performed using the BARDA Interactive Influenza Model. In Figure 6.38 below, the value
of any given parameter was set at its lowest level or its highest level to produce the mean numbers of
deaths globally for those two values across model runs for all values of all other parameters. All
parameters can be explored with this analysis simultaneously EXCEPT for prior/natural immunity in the

population because the estimated R0 of a disease is calculated given natural levels of immunity (that is,
the value of these two parameters are linked). We explore the ability of a pathogen to evade natural
immunity separately.
Figure 6.38. Sensitivity analysis of global deaths resulting from an outbreak of seasonal influenza. The width
of the boxes correspond to the median prediction of deaths if the value for that parameter is set to its lowest
and highest level (and all other values are allowed to vary). Hollow boxes show the parameter value range for
community control measures and wild type pathogens. Grey sections of the boxes show how increases in the
transmissibility or pathogenicity beyond wild type levels affect consequences.
Firstly, this analysis demonstrates the great variability possible within the wild type strains that exist. The
least deadly (non-attenuated) wild type strains are predicted to cause fifteen-fold fewer deaths globally
than the most deadly wild type strains (from 100,000 to 4 million deaths). The least transmissible wild
type strains (that have not circulated recently) are predicted to cause six-fold fewer deaths than the most
transmissible wild type strains (from 80,000 to 500,000 deaths). Also, how an outbreak with seasonal
influenza would influence global morbidity and mortality in the context of currently circulating strains is
unknown. An unresolved question (which likely depends on the biology of the virus released and its
similarity to currently circulating strains) is if the laboratory-associated outbreak would replace the annual
toll of seasonal influenza by supplanting circulating strains or would it add to this toll. That is, if a
laboratory-associated outbreak causes 300,000 deaths, would that be in addition to the several hundred
thousand deaths expected annually or replace those expected deaths? Clearly, if a laboratory accident
occurred with a wild type, circulating strain, the accident would simply mimic the commonplace
occurrence of travel associated spread of influenza.
This analysis demonstrates that enhancing the pathogenicity of a seasonal influenza strain increases the
number of global deaths resulting from an outbreak significantly, largely due to the fact that the case
fatality rate of unmodified seasonal influenza is very low. Increasing the case fatality rate from its highest
level observed in seasonal influenza to that of 1918 pandemic influenza (5%), increases deaths by more
than 10-fold.
Increasing the transmissibility of seasonal influenza also increases the number of global deaths
significantly, but to a lesser degree than increases in pathogenicity. Increasing the R0 from 1.4 to 2.2 can
double global deaths.

From this analysis vaccines and antivirals have little influence on the global outbreak because of poor
public health infrastructure and resource availability across most of the world. For this reason, the GoF
phenotype leading to the evasion of the protection afforded by vaccination or antivirals does not
significantly increase global consequences.
However, when the outbreak in North America is considered alone (Figure 6.39), vaccines and antivirals
can reduce the deaths by an order of magnitude. This result may be surprising because the outbreak is
with an unanticipated serotype so no effective vaccine would be available for months. For a GoF strain of
influenza to overcome protection caused by a vaccine made specifically in response to the outbreak this
strain is causing, it must be modified to overcome immunity cause by any vaccine, not just a vaccine
matched to its serotype. Although the GoF literature describes how to alter the antigenic properties of
influenza, no one has described an experiment that makes an influenza strain overcome protective
vaccination regardless of its serotype. For this reason, only some GoF experiments leading to the evasion
of induced immunity increase consequences.
Figure 6.39. Sensitivity analysis of deaths in North America resulting from an outbreak of seasonal influenza.
The width of the boxes correspond to the median prediction of deaths if the value for that parameter is set to
its lowest and highest level (and all other values are allowed to vary). Hollow boxes show the parameter value
range for community control measures and wild type pathogens. Grey sections of the boxes show how
increases in the transmissibility or pathogenicity beyond wild type levels affect consequences.
When considering North America alone, antivirals also can reduce deaths significantly, by about an order
of magnitude. Although in a typical seasonal influenza outbreak, only about 5% of patients receive
antivirals, federal caches of influenza antivirals could accommodate a much greater level of treatment and
so overall death rates could drop significantly (largely through the prevention of secondary infections
from those administered antivirals during treatment—compare the width of the bars for AV efficacy at
preventing transmission vs AV effectiveness on mortality). For this reason, resistance to antivirals in a
modified strain could increase the death toll of an influenza outbreak in the US by about an order of
magnitude (fourth and fifth box from top in Figure 6.39) even though this phenotype would have
negligible influence on the number of deaths globally (fourth and fifth box from top in Figure 6.38).

6.7.2 Pandemic Influenza Virus
Even if a wild type strain of pandemic influenza sparked a global outbreak the consequences would
eclipse those from all industrial accidents ever suffered. This section describes which GoF phenotypes
would influence the consequences of a global outbreak of pandemic influenza strains. The GoF
phenotypes relevant to an ongoing global outbreak are:
• The ability to overcome protective vaccination,
• The ability to overcome prior immunity (either natural or induced by previous vaccinations or
infection by similar strains in the past),
• Resistance to antivirals,
• Transmissibility, and
• Pathogenicity (used here, case fatality rate)
As above the ability to evade diagnostics is of secondary importance to the effect of antivirals and
vaccines and the GoF phenotypes of enhanced growth and adaptation to mammalian hosts are irrelevant
to an ongoing outbreak.
To understand how various GoF phenotypes influence the consequence to a global outbreak of pandemic
influenza, a sensitivity analysis was performed as described above (Figure 6.40). All parameters can be
explored with this analysis simultaneously EXCEPT for prior/natural immunity in the population because
the estimated R0 of a disease is calculated given natural levels of immunity (that is, the value of these two
parameters are linked). The ability of a pathogen to evade natural immunity is explored separately.
Figure 6.40. Sensitivity analysis of global deaths resulting from an outbreak of pandemic influenza. The width
of the boxes correspond to the median prediction of deaths if the value for that parameter is set to its lowest
and highest level (and all other values are allowed to vary). Hollow boxes show the parameter value range for
community control measures and wild type pathogens. Grey sections of the boxes show how increases in the

Firstly, this analysis demonstrates the great variability possible within the wild type strains that exist. The
least deadly (non-attenuated) wild type strains are predicted to cause 400-fold fewer deaths globally than
the most deadly wild type strains (from 200,000 to 80 million deaths). This result mirrors our previous
observations from the recent 2009 pandemic and the 1918 pandemic, which demonstrated an enormous
difference in their case fatality rate. The least transmissible wild type strains are predicted to cause ten-
fold fewer deaths than the most transmissible wild type strains (from 100,000 to 1,000,000 deaths).
From this analysis, we note that any GoF-related modification in pandemic influenza may not have a
significant effect on risk because these strains are already highly transmissible and pathogenic (more so
than seasonal strains). Increasing the case fatality rate beyond that of 1918 pandemic influenza (5%)
increases the number of deaths modestly. That being said, increasing the pathogenicity of an average
pandemic strain to that of the 1918 strain can increase the fatalities expected by 80-fold. Increasing
transmissibility increases the number of global deaths by at most a factor of three.
From this analysis vaccines and antivirals have little influence on the global outbreak because of poor
public health infrastructure and resource availability across the world. For this reason, the GoF phenotype
leading to the evasion of the protection afforded by vaccination or antivirals does not significantly
increase global consequences.
Even when North America is considered alone, for pandemic strains, vaccine evasion and antiviral
resistance influences potential deaths by little because the outbreak spreads quickly due to the increased
transmissibility of wild type pandemic strains compared to seasonal strains (Figure 6.41).
Figure 6.41. Sensitivity analysis of deaths in North America resulting from an outbreak of pandemic
influenza. The width of the boxes correspond to the median prediction of deaths if the value for that
parameter is set to its lowest and highest level (and all other values are allowed to vary). Hollow boxes show
the parameter value range for community control measures and wild type pathogens. Grey sections of the
boxes show how increases in the transmissibility or pathogenicity beyond wild type levels affect consequences.
6.7.3 Avian Influenza Virus That is Transmissible Amongst People
Wild type strains of avian influenza are unable to cause a global pandemic unless they are transmissible
amongst people. For this reason, the consequences of a global outbreak for GoF phenotypes other than

transmissibility all must be considered in the context of a strain that is already modified to be highly
transmissible. This interaction is explored below.
6.7.4 Coronaviruses
Even if a wild type strain of a SARS-like CoV sparked a global outbreak the consequences would eclipse
those from all industrial accidents ever suffered. This section describes which GoF phenotypes would
influence the consequences of a global outbreak caused by a SARS-like CoV. We focus on SARS-like
CoVs because wild type MERS-CoV is not sufficiently transmissible in people to cause a global
pandemic. If MERS-CoV were modified to be more transmissible, the resulting outbreak would resemble
that caused by a SARS-like CoV. The GoF phenotypes relevant to an ongoing global outbreak are simply
transmissibility and pathogenicity because there are no medical countermeasures to forestall the spread of
the illnesses caused by the coronaviruses.
As above the GoF phenotypes of enhanced growth and adaptation to mammalian hosts are irrelevant to an
ongoing outbreak (unless they alter transmissibility or pathogenicity). To understand how various GoF
phenotypes influence the consequence to a global outbreak of a SARS-like disease, a sensitivity analysis
was performed as described above (Figure 6.42). As the data show, increasing transmissibility of SARS-
CoV beyond wild type levels can double the prediction of median global deaths under certain
circumstances, a similar affect to increasing the pathogenicity. Of note, variation in the estimates of wild
type transmissibility of SARS-CoV can increase or decrease global deaths by 100,000-fold, showing how
little effect a modification can have compared to natural variation (or imperfect epidemiological
estimates). Similarly, the ability of the community to reduce their contacts for a significant period of time
has a similar influence on the consequences of the outbreak.
Figure 6.42 Sensitivity analysis of global deaths resulting from an outbreak of SARS. The width of the boxes
correspond to the median prediction of deaths if the value for that parameter is set to its lowest and highest
level (and all other values are allowed to vary). Hollow boxes show the parameter value range for community
control measures and wild type pathogens. Grey sections of the boxes show how increases in the
Firstly, this analysis demonstrates the variability possible within the wild type strains that exist. The least
deadly wild type strains are predicted to cause four-fold fewer deaths globally than the most deadly wild
type strains (from 200 to 800). The lowest estimates for the transmissibility of the coronaviruses (which,

as mentions above, are insufficiently transmissible to cause a global epidemic) result in outbreaks that
cause 100,000-fold fewer deaths than the outbreaks modeled with the highest estimates of
transmissibility. If the worst-case estimates for transmissibility for a SARS outbreak were used one could
expect a global outbreak to kill several million people. Recall that since SARS is very susceptible to
control measures, much of the difference in these estimates is likely due to the robustness of public health
measures undertaken to curtail its spread.
A global outbreak of a SARS-like disease would differ from a global influenza outbreak in many ways,
several of which are explicitly explored in this analysis (like case fatality rate, existence of medical
countermeasures, etc). Beyond these traits, SARS has a much longer incubation time (median of more
than four days but a much greater average) than influenza and therefore a global outbreak of SARS would
be much more protracted than an outbreak of influenza. Figure 6.43 shows when the peak number of
daily cases of a SARS-like disease compared to the initiation of the outbreak. For the smallest outbreaks,
the peak is reached within the first 500 days. However, other outbreaks require many years to reach their
peak in terms of cases per day. Clearly, the protracted nature of a SARS-like disease pandemic could put
a greater strain on sustaining a response and, conversely, afford some additional opportunities for
outbreak control compared to an influenza pandemic that circulates in less than a year. Given that an
outbreak of this kind has never been experienced, the nature and effect of these possibilities cannot be
quantified in the current modeling effort.
Figure 6.43. The number of coronavirus outbreaks modeled that peak (in terms of new cases per day) at any
particular day after the global outbreak begins. To show the duration of truly global outbreaks, outbreaks
that lead to less than one million infections are not shown.
The sections that follow provide a drill-down to describe HOW changes in any of the GoF phenotypes
affect the consequences of a global outbreak.
6.7.5 Effect of enhanced transmissibility in mammals on consequences of a global outbreak
As discussed above, increasing the transmissibility of a seasonal influenza strain can double the global
death toll. Figure 6.44 explores this relationship in more detail. These data show that increasing the

transmissibility of seasonal influenza to match that of an average pandemic influenza outbreak (R0 of 1.7)
is sufficient to double the death toll and increases beyond that point do no further increase consequences
significantly. For relatively poorly transmissible strains of seasonal influenza, increasing the
transmissibility to the greatest levels observed for wild type strains (in blue on the left in Figure 6.44) can
increase global deaths by 50 fold if no social distancing measures are taken during the outbreak. Recall
that community mitigation is a parameter that describes the actions taken by the public to reduce their
contacts with potentially infected individuals (such as avoiding public gatherings and mass transit).
Essentially, community mitigation reduces the ability of the disease to spread effectively in the
population.
Figure 6.44. Relationship between transmissibility of a seasonal influenza strain (in R 0 of the outbreak) and
global deaths. Grey points are used to show values for R0 beyond the estimates for wild type seasonal
influenza strains corresponding to the tornado plot in Figure 6.27. “Error bars” show the range of results
lowest 10% of results are not represented).
As discussed above, increasing the transmissibility of a pandemic influenza strain has little influence on
the global death toll. Figure 6.45 explores this relationship in more detail. These data show that increasing
the transmissibility of pandemic influenza has no effect unless a robust community mitigation effort can
be sustained throughout the outbreak (community mitigation 0.5 in the figure below). Given that these
outbreaks last many months, the ability for the community to sustain this level of social distancing is
doubtful especially given that this level of community mitigation has not been observed in any prior
modern influenza outbreak.

Figure 6.45. Relationship between transmissibility of a seasonal influenza strain (in R 0 of the outbreak) and
global deaths for various levels of sustained community mitigation. Grey point are used to show values for R0
beyond the estimates for wild type pandemic influenza strains and correspond to colors in Figure 6.29.
“Error bars” show the range of results across 80% of the parameter values explored in the analysis (the
parameter values that cause the highest and lowest 10% of results are not represented).
In Figure 6.46, we show the relationship between R0 and global cases of avian influenza for four levels of
community mitigation. With a novel, highly pathogenic illness, we can expect the public to significantly
change their behavior, as was observed in the SARS outbreak in Canada. However, we lack the data to
predict to what degree social distancing can be implemented and for what period of time. Figure 6.46
shows, however, that unless very significant levels of community mitigation can be sustained for a very
long time, the number of global cases significantly increases as the R0 of an avian influenza strain
approaches that of seasonal influenza. Increasing the R0 past 1.5 (which is typical for pandemic influenza
strains) has no further effect on consequences. If, however, community mitigation can be sustained at a
very high level, then to significantly increase global consequences, the avian influenza strain must be
more transmissible than any pandemic influenza strain ever observed. Because the ability of the
community to reduce their contacts for a significant period of time is dubious, we presume that the
increase of the transmissibility of avian influenza to that of seasonal influenza significantly drives the
potential consequences of an outbreak.

Figure 6.46. Relationship between global consequences (in term of illnesses) and R0 of a modified avian
influenza virus and the strength of community mitigation. “Error bars” show the range of results across 80%
of the parameter values explored in the analysis (the parameter values that cause the highest and lowest 10%
of results are not represented).
Figure 6.47 shows the relationship between R0 and global cases of a SARS-like CoV for four levels of
community mitigation. The data show that for wild type strains of SARS-like CoV, the virus is already
sufficiently transmissible (R0 > 1.4) to maximize global deaths unless very significant levels of
community mitigation can be sustained for a very long time. Figure 6.47 shows results for a SARS-like
CoV, but the results for a MERS-like CoV are nearly the same (not shown). That being said, because
MERS-CoV is less transmissible than SARS-CoV, a greater increase in transmissibility over a wild type
strain is required to produce the same increase in risk. If community mitigation can be sustained at a very
high level, then to significantly increase global consequences, the coronavirus strain must be more
transmissible than any influenza strain ever observed. Because the ability of the community to reduce
their contacts for the years required for a global outbreak to run its course is unknown, these data suggest
that SARS-like CoVs are already sufficiently transmissible to maximize a global outbreak and
modifications that increase transmissibility are of little additional risk.

Figure 6.47. Relationship between global consequences (in term of illnesses) and R 0 of a modified SARS-like
CoV and the strength of community mitigation. Strains modified to increase transmissibility beyond
estimates for wild type SARS-CoV are shown in grey. “Error bars” show the range of results across 80% of
the parameter values explored in the analysis (the parameter values that cause the highest and lowest 10% of
results are not represented).
6.7.6 Effect of enhanced pathogenicity on consequences of a global outbreak
Because of the low case fatality rate of typical seasonal influenza strains, increasing the pathogenicity of
these strains can significantly increase the predicted global death toll (Figure 6.48). These data show that,
as expected, increasing the case fatality rate by a factor of 10 or 100 increases the global deaths
correspondingly. An increase of this magnitude would be reflected by the modification of a typical
seasonal influenza strain to have the pathogenicity of the 1918 pandemic strain.

Figure 6.48. Relationship between global deaths and case fatality rate for seasonal influenza. Grey points are
used to show values for R0 beyond the estimates for wild type seasonal influenza strains corresponding to the
tornado plot in Figure 6.27. “Error bars” show the range of results across 80% of the parameter values
explored in the analysis (the parameter values that cause the highest and lowest 10% of results are not
represented).
Because the wild type 1918 pandemic influenza strain had a high case fatality rate, increasing this rate by
a factor of 100 is impossible. Figure 6.49 shows the effect on global deaths of doubling the case fatality
rate of a pandemic strain to be 10% (double that of the wild type 1918 strain). These data show that, as
expected, doubling the case fatality rate doubles the global deaths correspondingly. Death rates beyond
10% have been observed in avian influenza strains only and then only rarely.

Figure 6.49. Relationship between global deaths and case fatality rate for pandemic influenza. Grey points
are used to show values for R0 beyond the estimates for wild type seasonal influenza strains corresponding to
the tornado plot in Figure 6.27. Because the wild type 1918 pandemic strain has a case fatality rate of 5%,
much of this graph is occupied by data reflecting wild type strains. “Error bars” show the range of results
lowest 10% of results are not represented).
If a strain of avian influenza that is already modified to be highly transmissible in people was further
modified to be more pathogenic in people, the number of deaths is expected to increase. The natural range
of pathogenicity in wild type strains of avian influenza is immense, from those that produce no clinical
symptoms in humans to those that kill about half of those with recognized illness. For this reason, the
GoF study that increases risk most is one in which the already pathogenic strain is made more
transmissible while pathogenicity is maintained, not a GoF study in which pathogenicity is increased.
However, as shown in Figure 6.50, below, expected global deaths increase linearly with increases in
pathogenicity. Figure 6.50 also confirms how much an influence contagiousness has on consequences, as
increasing the transmissibility beyond an R0 of 1.1 increases deaths by at least 10,000 fold.

Figure 6.50. The relationship between case fatality rate (a measure of pathogenicity), transmissibility (in
terms of R0) and global consequences (in terms of deaths). “Error bars” show the range of results across 80%
of the parameter values explored in the analysis (the parameter values that cause the highest and lowest 10%
of results are not represented).
Modifications to increase the pathogenicity (in this case, the case fatality rate) of a SARS-like have the
expected outcome in terms of global deaths as shown in Figure 6.51, in that doubling or tripling the rate
of death doubles or triples global deaths.

Figure 6.51. Relationship between global deaths and the case fatality rate of a SARS-like CoV. Strains
modified to increase pathogenicity are shown in grey. “Error bars” show the range of results across 80% of
the parameter values explored in the analysis (the parameter values that cause the highest and lowest 10% of
6.7.7 Effect of countermeasures evasion on consequences of a global outbreak
If the strain involved in the laboratory released were able to overcome protective vaccination, the
outbreak in North America would cause up to four-fold more deaths (Figure 6.52—vaccine efficacy
dropping from 0.6 or 0.4 to 0). Recall that the state of the public health infrastructure in the majority of
the world is so parlous that vaccines do not appreciably affect global death rates for influenza, so this risk
is realized only by high income countries. Also, this risk is realized only if the outbreak is caused by a
strain of influenza that can overcome protective immunity afforded by any vaccine (instead of simply
changing the antigenic properties of the virus) because even if a novel strain has unprecedented antigenic
properties, a vaccine developed in the midst of an outbreak would be raised to the strain causing the
nascent outbreak. For these reasons, this GoF trait poses little overall biosafety risk.

Figure 6.52. Relationship between vaccine efficacy and deaths from a seasonal influenza outbreak in North
America. Because this outbreak is caused by a laboratory accident, the vaccine is raised to the strain involved
in the outbreak soon after the outbreak occurs. “Error bars” show the range of results across 80% of the
parameter values explored in the analysis (the parameter values that cause the highest and lowest 10% of
In a typical influenza seasons, only approximately five percent of patients receive antivirals. If this level
of antiviral distribution is used in the face of an outbreak caused by a laboratory accident, very few lives
are saved by antivirals and therefore antiviral resistance has limited influence on risk (Figure 6.53—
darker green line). However, the US holds a very large federal cache of antivirals that could be used to
provide treatment for many victims in a serious influenza epidemic. One may presume that if a global
outbreak were caused by an accident in a US laboratory, this cache would be deployed and used
aggressively. In this case, antivirals can significantly reduce risk of an outbreak by preventing the onward
transmission of influenza (Figure 6.53—light green line). Conversely, a seasonal influenza strain that is
antiviral resistant could vitiate the protection afforded to the public by antivirals and could increase the
consequences of an outbreak in North America by five-fold. We do not know how many other countries
have similar large caches of antivirals so we cannot determine if this risk increase would be shared by the
rest of the high income countries.

Figure 6.53. Relationship between antiviral efficacy and deaths from seasonal influenza in North America.
The left panel shows efficacy in terms of the ability to prevent mortality, the right is in terms of preventing
onward transmission. Typically, only about 5% of influenza patients receive antivirals. However, the US has
a large cache of antivirals that could be used in case of an emergency. Typical use and possible use are shown
in the graph. “Error bars” show the range of results across 80% of the parameter values explored in the
analysis (the parameter values that cause the highest and lowest 10% of results are not represented).
If the strain involved in the laboratory released were able to overcome protective vaccination, the
outbreak in North America would cause up to 25% more deaths (Figure 6.54—vaccine efficacy, defined
by the percent reduction in infection risk for a vaccinated individual compared to an unvaccinated
individual, dropping from 0.6 or 0.4 to 0). Recall that the state of the public health infrastructure in the
majority of the world is so parlous that vaccines do not appreciably affect global death rates for influenza,
so this risk is realized only by high income countries. Also, this risk is realized only if the outbreak is
caused by a strain of influenza that can overcome protective immunity afforded by any vaccine (instead of
simply changing the antigenic properties of the virus) because the vaccine would be raised to the strain
causing the nascent outbreak. For these reasons, this GoF trait poses little overall biosafety risk.

Figure 6.54. Relationship between vaccine efficacy and deaths from a pandemic influenza outbreak in North
America. Because this outbreak is caused by a laboratory accident, the vaccine is raised to the strain involved
in the outbreak soon after the outbreak occurs. “Error bars” show the range of results across 80% of the
parameter values explored in the analysis (the parameter values that cause the highest and lowest 10% of
If only 5% of the population receives antivirals, then risk is barely mitigated (Figure 6.55—darker red
line). If antivirals were more widely distributed, these countermeasures can reduce risk of an outbreak by
two to three fold by preventing the onward transmission of influenza and by preventing mortality (Figure
6.55—light red line). Conversely, a pandemic influenza strain that is antiviral resistant could vitiate the
protection afforded to the public by antivirals and could increase the consequences of an outbreak in
North America by two-to three-fold. We do not know how many other countries have similar large caches
of antivirals so we cannot determine if this risk increase would be shared by the rest of the high income
countries.

Figure 6.55 Relationship between antiviral efficacy and deaths from pandemic influenza in North America.
The left panel shows efficacy in terms of the ability to prevent mortality, the right is in terms of preventing
onward transmission. Typical use and possible use are shown in the graph. “Error bars” show the range of
results across 80% of the parameter values explored in the analysis (the parameter values that cause the
highest and lowest 10% of results are not represented).
An unexpected outbreak of avian influenza that is highly transmissible amongst people would be very
difficult to control with vaccination. The disease will have spread significantly by the time a protective
vaccine could be developed, tested, made in quantity and deployed. For this reason, as shown in Figure
6.56, below, the efficacy of the vaccine (and therefore, the ability of the pathogen to evade protective
vaccination), matters only for a narrow range of R0 values for avian strains. North America is shown
because this region has greater resources and capacity than the world as a whole so that vaccines can
show some efficacy. If the strain is highly transmissible, then vaccination comes too late to prevent a
significant number of deaths. If the strain is as transmissible as the least transmissible seasonal influenza
strains, then community mitigation is sufficient to contain the outbreak to a relatively low level without
vaccination.

Figure 6.56. Consequences (in terms of deaths) in North America of a pandemic caused by an avian influenza
strain modified to be highly transmissible amongst people as a function of vaccine efficacy and
transmissibility. “Error bars” show the range of results across 80% of the parameter values explored in the
analysis (the parameter values that cause the highest and lowest 10% of results are not represented). The
panel above shows a modest level of community mitigation sustained throughout the pandemic, and the panel
below shows a more robust level of community mitigation sustained through the pandemic.

Two critical points must be made. Firstly, because a vaccine will be developed for the serotype driving
the outbreak, the evasion of a vaccine is only relevant if the strain is modified to evade protection by any
vaccine, regardless of the antigenic properties of the virus. Secondly, predicting how transmissible an
avian strain could become is difficult. Although seasonal influenza strains are highly transmissible, an
avian strain could, theoretically, become as transmissible as a pandemic strain if much of the increase in
transmissibility of pandemic strains over seasonal strains is due to the lack of protective innate or residual
immunity in the population. For this reason, the value of protective vaccination against an unexpected
outbreak caused by an avian strain is very difficult to predict with certainty.
In summary, two facts significantly limit the risk posed by a transmissible avian strain of influenza that
can evade vaccination. Firstly, a narrow combination of phenotypes and control measures are necessary
for vaccines to have a significant effect on the outbreak even in North America (where the response
capacity is much greater than the world as a whole). Secondly, to affect risk at all, the strain must be able
to overcome protective vaccination regardless of the serotype of the virus. Although this modification
poses a biosecurity risk (see below), it is not the subject of active research (and also of dubious scientific
benefits) and so poses little biosafety risk.
Currently, no countermeasures specific to infections by the coronavirus are used to treat illnesses caused
by this pathogen or prevent the ongoing spread of an outbreak caused by this pathogen. For this reason,
this phenotype has no influence on risk.
6.7.8 Effect of evasion of natural/residual immunity on consequences of a global outbreak
As mentioned above, the baseline sensitivity analysis does not investigate the sensitivity to innate/residual
immunity in the population, which can be significant for seasonal influenza strains, because population
immunity is already accounted for in the effective R0. In Figure 6.57 below, we investigate how changes
in innate or residual immunity affects global deaths for a given R0 values.

Figure 6.57. The effect of the evasion of population immunity on global deaths from an outbreak of seasonal
influenza. The result for the baseline parameter value for prior immunity is the leftmost box in each panel. In
the left-hand panel, the baseline assumption is that 40% of the population is protected against infection with a
wild type strain. The graph shows the increase in the number of deaths when a strain (with the same R0
value) were able to overcome most of the immunity (so that only 10% of the population were immune) or
overcome all immunity. In the right-hand panel, the baseline assumption is that 10% of the population has
immunity to the wild type strain. The graph shows the increase in number of deaths deaths when the strain is
modified to overcome all immunity.
This analysis demonstrates that the evasion of pre-existing immunity can increase global deaths by nearly
four-fold, if the population has a high level of residual immunity (as is likely for seasonal influenza since
prior vaccination or illness provides some protection against new strains). Similar to R0, this parameter
influences the global outbreak by enabling the disease to spread more quickly (because each contact is
more likely to result in an infection) and eventually infect a larger number of people worldwide. Pre-
existing immunity can protect a significant proportion of the population if the strain released is similar (or
identical) to a strain of influenza that recently circulated, which is one reason why this parameter is
influential. If the population exhibited relatively low levels of prior immunity, then evasion of prior
immunity has little influence on consequences (increasing global deaths by less than two-fold).
As mentioned above, the baseline sensitivity analysis does not investigate the sensitivity to innate/residual
immunity in the population because population immunity is already accounted for in the effective R0.
Given the evolution of pandemic strains, residual immunity is expected to be minimal, which probably
accounts for the observed greater R0 of pandemic strains than seasonal strains. In Figure 6.58 below, we
investigate how changes in innate or residual immunity affects global deaths for a given R0 values.

Figure 6.58. The effect of the evasion of population immunity on global deaths from an outbreak of pandemic
influenza. The baseline parameter value (left box) is 10% of the population is protected against infection with
a wild type strain. The graph shows how the number of deaths increase if a strain (with the same R 0 value)
were able to overcome all immunity.
This analysis demonstrates that the evasion of pre-existing immunity can modestly increase global
consequences for pandemic strains, partially because the level of prior immunity is considered to be low.
Also, because this parameter affects a global outbreak similar to R0 and increasing R0 changes global
consequences from pandemic strains only modestly, this result is not surprising.
Exposure to avian influenza strains is so rare in human populations that very few people have residual
immunity to this pathogen. For this reason, evasion of residual immunity has no influence on risk.
Exposure to a coronavirus is so rare in human populations that very few people have residual immunity to
this pathogen. For this reason, evasion of residual immunity has no influence on risk.
6.8 Supporting an Estimate of Absolute Risk
This assessment was designed to evaluate the increase in risk caused by the creation of strains of
pathogens with GoF traits compared to wild type pathogens. This approach enabled the assessment to
capitalize on the strengths of the available data and minimize the importance of the weaknesses.
Sufficient biomedical and epidemiological evidence exists to develop robust models of the initiation of an
outbreak from the primary to the secondary cases and the expansion of this outbreak within a community
to eventually spark a global pandemic. In contrast, very little data exists on human reliability in life
science laboratories, which drives the probability that laboratory acquired infections occur in the first
place. Fortunately, the accidents that humans cause (or contribute to) in the laboratory are the same
regardless of the pathogen manipulated. That is, workers may overfill a centrifuge tube with the same
frequency regardless of the pathogen in the tube, or will slip while working with scissors during a

necropsy with the same frequency regardless of the pathogen studied. Because the absolute rate at which
these accidents happen and cause infections is not supported by robust data, absolute estimates of the rate
of laboratory acquired infections cannot be made using the method described in this report.
However, to provide a context for the increase in risk suffered, absolute risk estimates are desired. For
this reason, the historical rate of laboratory acquired infections could be used to predict a reasonable
upper bound for the frequency with which these incidents occur. However, the research team is unaware
of any laboratory acquired infections in laboratories that study influenza or coronaviruses and so an
absolute risk analysis will have at its foundation a weak estimate of the frequency at which laboratory
acquired infections occur. That being said, this historical rate of laboratory infections can then be
combined with calculated rates of laboratory acquired infections leading to secondary infections, local
outbreaks and global pandemics from this assessment to produce an estimate of absolute risk.
The return frequency of laboratory acquired infections (LAIs) was estimated for several hypothetical
historical LAI counts, presuming that some historical LAIs may have gone undetected or unreported. LAI
frequency was modeled using a binomial distribution, with the number of trials set to the number of
laboratory-years (i.e., the number of laboratories working with the viruses times the observation period),
and the number of “successes” equal to the number of LAIs. For influenza, 100 labs359 and an observation
period of twenty years (for a total of 2,000 lab-years) was assumed because there has been roughly 20
years since the expansion of the life science research in the mid-1990s. This time period also coincides
with a wave of construction of modern biocontainment facilities that better represent the safety conditions
of today’s laboratories than previous laboratories.
Shown are the limits of the two-sided 80% confidence interval on the expected LAI frequency, estimated
using a Clopper-Pearson interval.360 The maximum likelihood estimate (MLE) of the frequency with
which an LAI would occur (the return frequency) was computed as the number of laboratory years
divided by the number of LAIs. Note that, for zero observed LAIs, the minimum and MLE return rate
approach infinity and are not plotted.
The project team knows of no laboratory acquired infections involving any one of these laboratories.361
This lack of a laboratory acquired infection could be due to the fact that none have occurred in that time
frame or that some have occurred but the project team does not have access to the reports or data. Figure
6.59 shows the limits of the 90% confidence interval (90 out of 100 times, LAIs would happen, on
average, less often) and maximum likelihood estimate of the return period of laboratory acquired
infections given that 0-10 infections have occurred in the past 20 years in the approximately 100
laboratories.
Across all 100 laboratories, a laboratory acquired infection could be expected as frequently as once every
8.5 years (if no infections have occurred in the last 20 years) to as little as every 200 years (if one
infection occurred). If the assumption is made that three LAIs have surreptitiously occurred, then an LAI
is expected to occur from once every three years to once every 20 years.
359 The exact number of laboratories does not significantly influence absolute risk (the per-laboratory rate decreases but the
absolute rate of an accident across all laboratories does not change).
360 Newcombe RG (1998) Two-sided confidence intervals for the single proportion: comparison of seven methods. Statistics in
medicine 17: 857-872
361 At least one parenteral exposure to H5N1 has occurred, and this worker was isolated, but never became ill, probably because
influenza is a respiratory pathogen and cannot infect muscle.

Figure 6.59. The predicted return period of laboratory acquired infections assuming 0-10 infections have
actually occurred in the last 20 years across 100 laboratories. The limits of the one-sided 90% confidence
interval of the maximum rate (bottom line) was used to produce an estimate of the return period that would
be greater than 90 out of 100 actual values of the frequency given the observations, whereas the maximum
likelihood estimate and limits of the one-sided 90% confidence interval of the minimum rate (top line) are
also shown.
The quantitative analysis in this report estimates that a small minority of these infections would start a
local outbreak and a minority of these outbreaks would seed a global pandemic.
For seasonal influenza, the analysis presented above suggests that only 0.4% of LAIs with seasonal
influenza are predicted to cause a global pandemic (assuming the strain has not recently circulated, in
which case, the probability would be even less). Because most of the 100 laboratories working on the
pathogens assessed in this report are studying seasonal influenza, this analysis suggests that a global
pandemic would be caused by a laboratory accident in the US once every 2,000-50,000 years (if
essentially no LAIs have occurred in influenza laboratories in the past 20 years). If instead the assumption
is made that three LAIs have surreptitiously occurred, a global pandemic could be triggered once every
750-5,000 years. It is worthy to note that viruses were characterized much less than 750 years ago, so it
cannot be stated with any certainty that these pathogens will be studied under similar containment
conditions for long enough into the future for an accident to be likely to occur even once. Moreover, the

true consequence of a seasonal influenza outbreak caused by a laboratory accident is unclear. Although
predictions can be made about the illnesses and deaths that would be caused, it is unknown how this
outbreak would influence the evolution and spread of other influenza strains and if these laboratory-
associated infections would supplant or supplement those expected on an annual basis. This caveat aside,
the analysis predicts 100,000-4,000,000 deaths to occur from a global outbreak of a wild type seasonal
influenza strain, depending on the pathogenicity and transmissibility of the strain.
Considering the other influenza viruses in this study, a historical analysis predicts that LAIs would occur
with a similar frequency assuming that no infections have occurred (the per-laboratory rate of LAIs
increases but fewer laboratories study these pathogens). This result is obviously counterfactual because
these pathogens are manipulated at a greater containment level than wild type, seasonal influenza viruses,
to decrease the probability of a LAI. That being said, a conservative estimate predicts that laboratory
acquired infections occur at the same rate as for seasonal influenza viruses. The analysis presented above
suggest that only 1.5% of LAIs with pandemic influenza are predicted to lead to global outbreaks.
Combined with the predicted return frequency of LAIs given no LAIs in the last 20 years, a global
pandemic caused by research on pandemic influenza viruses is expected every 560-13,000 years.
Assuming that no LAIs have occurred with the deadliest pandemic strains is reasonable because, it would
be widely known if several laboratory accidents occurred. An accident that sparks an epidemic with a
strain like the 1918 pandemic strain could cause up to 80 million global deaths according to the analysis
presented above. If, conversely, the accident occurred with a strain similar to the 2009 pandemic strain, it
would resemble an accident with seasonal influenza. Also, the continued circulation of similar influenza
strains and the recent exposure of many people to the 2009 strain suggests that residual immunity would
be unlikely to spread globally or cause a significant number of casualties. If, however, the strain were
antigenically distinct from the 2009 strain but were otherwise similar, roughly 200,000 deaths would be
expected to arise from a global pandemic.
Wild type avian influenza strains are not transmissible enough among people to cause a significant local
outbreak and therefore no global outbreak is possible. Assuming the same return frequency of laboratory
acquired infections for avian influenza as predicted for seasonal influenza, a laboratory worker is
expected to fall ill once every three to nine years. For the most pathogenic strains, this worker has a
significant chance of dying but the outbreak is likely to extend no further than that one case.
Given that SARS-CoV has been studied for only a decade, the historical record of no laboratory accidents
once again suggests that LAIs occur more frequently in coronavirus laboratories at BSL3 than in
laboratories that study seasonal influenza at BSL2, which is obviously wrong. If, conservatively, the
estimate is made that LAIs with SARS occur as frequently as influenza, a LAI is expected to occur once
every 8.5 years (given the seriousness of SARS, LAIs are likely to have been reported so it is safe to
assume that no LAIs have occurred yet with this pathogen in the US). The best estimates for the
transmissibility of SARS suggest that there is no chance that this outbreak would spark a global pandemic
(and SARS-CoV is more transmissible than MERS-CoV). Most of these LAIs would lead to no further
infections, however, some would lead to the infection of a handful of other individuals.

6.9 Using the Parametric Risk Assessment: Example Calculation
By design, the biosafety risk assessment is broad and provides data to understand how risk changes if a
wild-type pathogen is manipulated in one of a variety of ways. This section provides an example
illustrating how to simply use the information contained in this report to assess the risk posed by a
particular manipulation. This example compares the risk of research on two possible modified strains to a
wild type strain of influenza (called Strain 1). The example assumes that the strain has not circulated
recently so that it itself has some real biosafety risk. This example will use parameter values typical for a
wild type seasonal influenza strain as a baseline, which is described by the following parameters:
• Transmissibility: R0=1.3,
• Pathogenicity: Case fatality rate of 0.001,
• Antiviral sensitivity: (efficacy at preventing transmission=0.25, efficacy at preventing death=0.4),
• Vaccine protection: (efficacy of 0.5 at preventing infection), and
• Infectivity: Set to seasonal influenza (ID50 less than 10pfu)
This example uses two modified strains. Strain 2 is a GoF strain of seasonal influenza that is exactly like
the wild type strain, except that it is as transmissible as a strain pandemic influenza (R0=1.7). Strain 3 is
an attenuated strain of seasonal influenza that is exactly like the wild type strain, except that it has a case
fatality rate of 0.0001.
The modeling completed enables a complete assessment of how any combination of parameter values that
describe the pathogen and control measures influences risk, however, all possible combinations of these
values and their influence on risk cannot be shown concisely in a report. Instead, static slices through this
very complex risk space are taken and shown as two-dimensional figures in this report that explore the
effect of changing one parameter while allowing all others to vary. That is, using this report, phenotypes
must be assessed individually. The reader will note that the baseline results change regarding which trait
is being considered for the same strain (that is, which parameter is held at a particular value while all
others are allowed to vary). This phenomenon is expected because the figures show the median and 80th
percentile results for all of the parameters that COULD describe a wild type strain and a modified strain
(and the same range of values for the control measures). This parameter range will obviously change
depending on which trait is held constant.
6.9.1 Step 1: Determine if the probability of the pathogen escaping the laboratory changes
To determine how the probability of a pathogen escaping the laboratory changes as a pathogen is
modified, refer to section 6.4.4, specifically 6.4.4.1 for seasonal influenza. Table 6.3 shows how any trait
affects this probability. Neither of the traits considered in this example influence the probability that a
laboratory incident would lead to escape of the pathogen from the laboratory.
Recall that, although the analysis developed for this study will not permit an estimation of how frequently
laboratory accidents lead to laboratory acquired infections that spark a local outbreak, historical rates of
accidents suggest that a local outbreak would be sparked by a laboratory acquired infection about once
every 500-10,000 years.
6.9.2 Step 2: Determine the change in the probability of a resulting outbreak escaping local control
To determine the change in the probability that an outbreak, caused by a laboratory accident, would
escape local control and seed a global pandemic, refer to Section 6.6. This section demonstrates that of

the modifications described in this example, only transmissibility is known to have a significant influence
on the probability that an outbreak would escape local control. To determine HOW changes in this trait
affect this probability, refer to Section 6.6.1 and Figure 6.29, specifically, for seasonal influenza strains.
This figure shows that, should an outbreak resulting from an accident occur, this wild type strain (R0=1.3)
has a baseline chance of escaping local of roughly 21% (12-30% using 80% of the parameter values in the
assessment). If the highly transmissible Strain 2 (R0=1.7) were to cause a local outbreak, the probability
that the outbreak would escape local control and seed a global pandemic increase to 44% (27-50%).
Because, as described in Section 6.2.9 and Figure 6.12, the range of results in the figures reflect
monotonic increases with the same overall shape as the median estimate, one can directly compare the
median estimate and the extremes of the range given for any set of results. For this reason, this specific
modification is estimated to increase the probability that an outbreak would escape local control by 2.1-
fold (1.6-2.3x). If the transmissibility could be increased to rival the most transmissible pandemic
influenza strains, the risk of local escape would be further increased.
Table 6.8. Summary of the influence of exemplar modifications to seasonal influenza

on the probability that an outbreak, caused by a laboratory accident, would escape
local control and seed a global pandemic.
2=highly
Strain 1=wild type 3=attenuated
transmissible
Increase in
probability of an 1, no change due to
1, defined 2.1x (1.6-2.3x)
outbreak escaping modification
local control
6.9.3 Step 3: Determine if the consequences of a resulting pandemic changes
To determine how the consequences of a global pandemic changes, refer to section 6.7. Within section
6.7 refer to the section describing each modified trait of interest to understand how changes in that trait
affect this probability. This section demonstrates that both of the modifications described in this example
affect the consequences of a global pandemic.
To determine how changes in transmissibility affect consequence, refer to section 6.7.5.1 and Figure 6.44,
specifically, for seasonal influenza strains. This figure shows that, should a global pandemic occur, this
wild type strain (R0=1.3) would lead to roughly 500,000 global deaths (150,000-4,000,000 using 80% of
the parameter values in the assessment and assuming no community mitigation occurs). If the highly
transmissible Strain 2 (R0=1.8) were to cause a global pandemic, the deaths suffered would increase to
900,000 (250,000-10,000,000). Because, as described in Section 6.29 and Figure 6.12, the range of results
shown in the figures reflect monotonic increases with the same overall shape as the median estimate, one
can directly compare the median estimate and the high and low parts of the range given for any set of
results. For this reason, this specific modification is estimated to increase the consequences of a global
pandemic by 1.8-fold (1.6-2.5x).
To determine how modified pathogenicity affects global consequences should a global pandemic occur,
refer to section 6.7.6 and Figure 6.48, specifically for seasonal influenza. Although this figure visually
displays how case fatality rate affects global deaths, the change in deaths is simple to calculate as a 10-
fold decrease in the case fatality rate simply leads to a 10-fold decrease in global deaths.

Table 6.9. Summary of the influence of exemplar modifications to seasonal influenza
on the consequences of a global pandemic.
2=highly
Strain 1=wild type 3=attenuated
transmissible
Increase in global
1, defined 1.8x (1.6-2.5x) 0.1
consequences
6.9.4 Putting it together
Because the risk of a pandemic in this study is the product of the frequency of a laboratory incident
sparking a local outbreak, the frequency of an outbreak escaping local control and the consequences of a
global pandemic, the total change in risk can be simply understood as the product of the increases in any
of these values over the baseline.
The highly transmissible Strain 2 is as likely to escape from a laboratory, but 2.1-fold (1.6-2.3x) more
likely to cause a pandemic that would kill 1.8-fold (1.6-2.5x) more people than the wild type strain. In
total then, research on this strain poses 3.8-fold (2.6-5.8x) more risk of pandemics than research on a wild
type seasonal influenza strain. Put another way, GoF experiments that increase the transmissibility of
seasonal influenza to the level of pandemic influenza strains are 3.8-fold more risky than alternate
experiments involving wild-type strains. In contrast, if the research could be conducted with an attenuated
strain instead of a wild type strain, risk would decrease by a further 10-fold.

7 Biosecurity Risk of Malicious Acts Targeting a GoF
Laboratory
7.1 Biosecurity Risk Assessment: Summary 169
7.1.1 Malicious Actors and Acts 169
7.1.2 Security Governance 169
7.1.3 Qualitative Assessment: Plausible Threats 170
7.1.4 Conclusions 170
7.2 Findings: Assessment of the Offense (Possible Threats to U.S. Research Laboratories) 171
7.2.1 Malicious actors 173
7.2.2 Malicious acts 175
7.2.3 Detailed Descriptions 177
7.3 Findings: Defense Assessment 178

7.3.1 Overview of Security Measures 178
7.3.2 Detailed Descriptions 181
7.4 Analysis of Offense and Defensive Measures 181

7.4.1 Qualitative Assessment of Plausible Threats 182
7.4.2 Semi-Quantitative Epidemiological Modeling of Security Risks 199

7.1 Biosecurity Risk Assessment: Summary
The purpose of the biosecurity risk assessment is to provide NSABB with an assessment of the likelihood
that a malicious act involving a GoF influenza, SARS-CoV, or MERS-CoV virus could result in local
infections or widespread pandemic. The risk assessment involved five steps: 1) characterization of the
threat, which includes an evaluation of historical incidents and malicious actor motivation and capability
(the “offense”); 2) review of the current security policies and practices landscape that governs research
with influenza, SARS-CoV, and MERS-CoV in the United States (the “defense”); 3) identification of
plausible threats based on analysis of the “offense” and “defense”; 4) assessment of the potential for the
plausible threats to cause infections in the local community or broader; and 5) comparison of possible
pandemic consequences of plausible threats involving GoF viruses and non-GoF viruses.
influenza, SARS-CoV, or MERS-CoV virus is available. Therefore, to identify the types of actors and
acts that may target a GoF laboratory, our approach involved examining historical incidents involving life
science laboratories and hospitals, evaluating the motivations and capabilities of malicious actors, and
determining if and how existing security measures affect the likelihood of success of a malicious act.
Plausible threats facing laboratories that study or store GoF virus(s) were extrapolated from this
assessment. Figure 7.1 presents a schematic of the biosecurity risk assessment process.
Characterization of the Threat Governance Landscape

(“Offensive”) (“Defense”)
Identification of Plausible Threats

(Offense vs. Defense)
Assessment of Infection or Pandemic

Potential of Plausible Threats
Figure 7.1. Schematic of Biosecurity Risk Assessment Process. Detailed methodology is in Appendix V, Sec.
16.2 and 16.3.
7.1.1 Malicious Actors and Acts
In today’s regulatory and security environment, the main plausible threat facing high containment,
research laboratories that store or study GoF viruses, involves malicious insiders who have authorized
access to the laboratories and virus(s) contained therein. Insiders may work alone or in coordination with
an outside group. Their motivations range from emotional disturbances to ideological radicalization by
domestic and transnational terrorist organizations. The likelihood that outsiders could gain access to a
laboratory without insider assistance is low. Therefore, outsiders present a threat to the periphery of the
research complex or building only, but not a significant threat to the high containment laboratory itself.
7.1.2 Security Governance
Governance of infectious disease research is extremely complex, involving international agreements,

domestic law, guidance, and contractual requirements, in addition to institutional, local, and state-specific
policies. Highly pathogenic avian influenza, the reconstructed 1918 influenza, and SARS-CoV viruses are

all Select Agents and are therefore covered by the Select Agent Regulations. Low pathogenic avian
influenza and MERS-CoV are not Select Agents. Security systems, protocols, and practices at non-select
agent, select agent, and Tier 1 select agent levels were reviewed to evaluate the likelihood of a malicious
actor carrying out a successful act involving a laboratory that stores or studies a GoF virus. Analysis of
plausible threats accounts for current security measures at the lowest level at which GoF research is
conducted.
7.1.3 Qualitative Assessment: Plausible Threats
Based on historical incidents, the most likely malicious acts to be carried out in or on a laboratory that
studies or stores GoF virus(s) include removal of the virus from frozen stocks, experimental samples,
equipment, or research animals; deliberate contamination of personal protective equipment or laboratory
equipment; deliberate compromise of the personal protective equipment or laboratory equipment; and
mixing of infected with uninfected samples or animals outside proper containment. In addition, incidents
involving bombs or active shooters may cause loss of containment if carried out inside or near the
entrance of high containment laboratories in which GoF research is conducted. Noncompliance with
security regulations and networked control systems might increase laboratory biosecurity risks.
Table 7.1 summarizes these plausible threats, including both malicious actor and act.
Table 7.1. Plausible Threats Involving High Containment Research Laboratories that Store or Study GoF
Viruses
Active shooter or physical assault
Insider
Overt Bomb detonated near or inside high containment space
Outsider Bomb detonated at building periphery
Removal of GoF virus (frozen stock or experimental sample),
Covert Act (Expose Public) Insider
infected animals, or contaminated equipment
Removal of GoF virus in experimental samples
Deliberate contamination of personal protective equipment or
Covert Act (Expose laboratory equipment
Insider
Laboratory Workers) Deliberate compromise of laboratory equipment or personal
protective equipment
Mixing of experimental samples or animals into lower containment
7.1.4 Conclusions
The existing regulatory infrastructure governing influenza, SARS-CoV, and MERS-CoV appears to
provide sufficient defenses, if properly implemented, against unauthorized outsiders from accessing
modified viruses. However, clarity and guidance associated with current policies could be improved to
enhance compliance with required security measures. In addition, data that could be used to inform the
need for additional security measures does not exist (or is not in the public domain).
Only a handful of GoF traits significantly increase biosecurity risk after a malicious event targets a
laboratory. For seasonal and pandemic influenza, the ability to overcome protective vaccination and
antiviral resistance modestly increases risk by increasing the potential consequences in the high income
countries. There is no significant effect on risk if the global population is considered as a whole. No other
trait influences the risk significantly of pandemic influenza strains. For seasonal influenza, increasing the
transmissibility and ability to evade residual immunity significantly increases risk because outbreaks are

more likely to occur, to escape local control and will create more consequential global outbreaks. For
avian influenza, increasing transmissibility greatly increases risk because this modification is required to
spark a global outbreak of a disease by human-to-human contact, potentially infecting millions. Without
this change, the hazard is restricted to those exposed to contaminated materials and infected birds,
limiting the outbreak to thousands of cases at most. Increasing pathogenicity can modestly increase risk.
When comparing the biosafety and biosecurity risks, a successful event that covertly infects the public
(theft from an influenza laboratory of an infected animal, contaminated piece of equipment or viral stock)
must occur once every 65-190 years for biosecurity event to have the same total risk as biosafety events.
Given the frequency with which these malicious acts have occurred in the past, this analysis suggests that
7.2 Findings: Assessment of the Offense (Possible Threats to U.S. Research

Laboratories)
Incidents of criminal, terrorist, and illicit governmental activities involving pathogens, U.S. laboratories,
and/or researchers have been documented in several books, articles, official government documents, and
other open source publications. However, the potential risks of intentional or accidental release of
laboratory-generated or adapted pathogens into the community or environment from deliberate acts whose
main goal is not bioterrorism often are not included in these accounts. Similarly, the risk that cyber
breaches may result in the intentional disruption of facility operations has not been fully described in open
literature. The lack of publicly available data about the likelihood that a cybersecurity breach could
disrupt facility operations and control systems of high containment laboratories makes assessing such
threats prohibitively difficult in unclassified settings. Therefore, the potential threats to human health that
cyber breaches pose are not addressed in this report.
In our assessment of the potential biosecurity threat associated with GoF influenza, SARS-CoV, and
MERS-CoV research, a variety of malicious actors, malicious acts, and consequences, including
deliberate incidents that resulted in accidental release and cyber security breaches were evaluated.
Furthermore, the motivations and capabilities of each malicious actor type based on conventional
knowledge and historical events found in open source documents were evaluated. Finally, historical
incidents of deliberate harm or application of science for destructive purposes were considered in this
analysis.
The following section summarizes actual malicious acts against laboratories and health care facilities, or
involving attempts to acquire pathogens based on analysis of the open-source, historical literature. Figures
7.2 – 7.4 summarize all of the historical events that occurred in the United States over the past 25 years
based on open source reporting.

Figure 7.2. Historical acts malicious actors carried out in the United States. The black circles indicate two or
more historical events, the grey circles indicate one historical event, and the cross-hatched black-grey circles
indicate one or two historical events.
Figure 7.3. Consequences resulting from historical malicious acts carried out in the United States. The black
circles indicate two or more historical events and the grey circles indicate one historical event.

Figure 7.4. Possible consequences resulting from acts carried out in the United States by malicious actors.
The black circles indicate two or more historical events, the grey circles indicate one historical event, the grey
hatched circles indicate group efforts that may/may not have been associated with a criminal organization,
and the blue circles indicate hopeful outcomes of planned or attempted (but failed) events. The “R” in one of
the black bubbles indicates the Rajneeshee Cult, who are the only group of domestic terrorists that
deliberately exposed members of the public.
7.2.1 Malicious actors
The following key findings are reached regarding malicious actors, based on the research presented in the
section above:
• Analysis of past acts by malicious actors that involved a U.S. laboratory shows that, out of a
universe of possible acts, relatively few malicious acts with the potential to lead to a breach in
containment have been carried out.
• The majority of the documented prior acts have been committed by domestic terrorist and
extremist groups, specifically by animal rights extremists. These groups have engaged in a wide
range of malicious acts, including laboratory arson, sabotage, subversion of employees, and
reckless acts such as the release of laboratory animals. Although these actions probably did not
seek the release of a pathogen from a laboratory, they nevertheless have resulted in such an
outcome on at least one occasion. In one case dating from 1989, animal rights extremists released
30 infected mice infected with cryptosporidium from a laboratory, probably without knowing that
the mice were infected.362 However, animal rights extremists have been rigorously pursued and
362 “Diseased mice freed in arson fires, break-in,” Spartanburg Herald-Journal, April 4, 1989, A2. Retrieved at:
https://news.google.com/newspapers?nid=1876&dat=19890404&id=2kwsAAAAIBAJ&sjid=Vs4EAAAAIBAJ&pg=6664,1
859692&hl=en.

arrested by FBI in recent years.363,364,365,366 The number of animal rights attacks appears to have
decreased in recent years because of increased security at research institutions367 and increased
arrests by law enforcement.
• Many documented events occurred before new counter-terrorism and counter-extremist laws and
policies were put into place. With these new requirements in place, carrying out a malicious act
today against a high containment laboratory in the United States is challenging, which might
deter or prevent groups from repeating past attacks. This idea was highlighted in propaganda
from one animal rights extremist group, which blamed their decreased activity against
laboratories on the difficulty of penetrating “increased security.”368
• Insiders pose a significant risk because of lone actor incidents, the unpredictability of emotionally
disturbed insiders, potential for radicalization by extremists or terrorists, or elicitation or
subversion incidents. Insiders have carried out or been involved in malicious acts involving the
diversion of a pathogen from a laboratory and the infection of someone in the general public. In
addition, noncompliance with security regulations increases the potential biosecurity risk posed
by insiders.
• Transnational terrorist groups, including state-like groups, were found to be unlikely to target
U.S. laboratories directly through armed assaults or bombings. However, foreign terrorist
organizations, such as al Qaeda and ISIL, have issued calls for scientists, doctors, and engineers
to join their cause, which includes the use of specialized skills to inflict harm.
• Foreign intelligence entities have and continue to target biological laboratories to steal
information or laboratory materials. These efforts can be done through elicitation or subversion of
laboratory employees, insertion of an operative, or more recently, remotely through hacking into
institutional computer networks. No information in open source literature links these incidents of
theft to release of a biological agent.
• In the past, a select few foreign intelligence agencies weaponized biological agents for use in
assassinations. In addition, the Soviet Union’s KGB targeted Western research on modified
pathogen strains, possibly to bolster the Soviet offensive program.369
363 John E. Lewis, Deputy Assistant Director, Federal Bureau of Investigation, Testimony before the Senate Judiciary
Committee, Washington DC., U.S.A., May 18, 2004, https://www.fbi.gov/news/testimony/animal-rights-extremism-and-
ecoterrorism.
364 Moran R, “Animal activists defend tactics that led to raid – Protests target the homes of business executives,” The Inquirer,
November 22, 2004, http://articles.philly.com/2004-11-22/news/25379045_1_huntingdon-life-sciences-animal-activists-
animal-rights.
365 Law enforcement efforts outside of the U.S. have also targeted animal rights extremists in recent years. Mark Oliver, “30
arrested as raids target animal rights extremists,” The Guardian, May 1, 2007,
http://www.theguardian.com/uk/2007/may/01/animalwelfare.world.
366 Patrick Sawer, “Debbie Vincent: Former soldier turned animal rights extremist jailed for six years,” The Telegraph, April
17, 2014, http://www.telegraph.co.uk/news/uknews/crime/10772486/Debbie-Vincent-Former-soldier-turned-animal-rights-
extremist-jailed-for-six-years.html.
367 The following extracts from a website maintained in support of the Animal Liberation Front, a domestic extremist animal
rights organization, supports this claim: “Numerous larger liberations took place in the early eighties before technologically
advanced security systems were placed in most larger animal laboratories” and, “Because of increased security, liberations
haven't been as frequent in the 1990's […]” “Laboratory Animal Liberation Campaign,” Animal Liberation Front,
http://www.animalliberationfront.com/ALFront/lab.htm.
368 Ibid.
369 Leitenberg M., Zilinskas R., (2012) The Soviet Biological Weapons Program: A History. Cambridge, MA, Harvard
University Press

• Organized criminals have not attempted to steal pathogens from a U.S. laboratory. However, one
case of cyber-crime suggests that theft of information on applied life science research can be
lucrative, and therefore tempting, for organized criminal groups.
• Lone outsiders do not pose a significant threat to research laboratories, especially Biological
Select Agent and Toxin laboratories, because they do not have access to the facilities. By
definition, these actors are not working with an insider and would not have opportunities to gain
access to facilities in the absence of intentional or unintentional assistance370 of an insider.
Interviews confirmed the following threats of concern:
• Insiders with access to information and pathogens and who become discontented or disgruntled,
radicalized, or elicited or subverted are a security concern.
• Transnational terrorists who are interested in biological weapons are a security concern.
• Domestic extremists, such as animal rights extremists, anti-vaccine extremists, and eco-radical
groups, who see harming researchers and institutional administrators, and/or vandalizing
institutional facilities as a useful approach to convey their messages are a security concern The
threat posed by domestic extremists appears to vary by the laboratory’s location.
• Lone outsiders do not raise much concern because they are not working with an insider and have
difficulty accessing laboratories and breaching facility defenses unassisted.
• Active shooters on university campuses are of significant concern even though no incidents
involving an active shooter in a high containment laboratory have been described.
7.2.2 Malicious acts
The following key findings are reached regarding malicious acts, based on the research presented in the
section above:
• Armed assaults at laboratories have not occurred previously, but with increasing incidents of
active shooter cases on university campuses, the potential for armed assault might exist.
• A bombing attack against a U.S. lab has not taken place. Because malicous actors have bombed
hospials and related facilities, this type of attack remains possible.
• No exposure or release events have occurred from sabotage, despite the relative frequency of
sabotage incidents.
• Reckless acts provide the greatest opportunity for an outbreak to occur, and have occurred on
several occasions in the past, as documented above.
• A deliberate infection of a member of the public through a deliberate or reckless act by an insider
represented the most common pathway for loss of containment across the spectrum of malicious
actors and acts considered.
370 An example of unintentional assistance by an insider is access through an insider not complying with physical, information,
or transportation security measures.

• A reckless act involving the release of an infected animal outside of containment has occurred
once before, as a result of an attack by a domestic animal rights extremist group who likely did
not know the mice were infected.
• Deliberate self-infection remains a hypothetical concern. The closest event documented in open
source literature is one reported HIV self-infection case involving an outsider without lab access
who attempted suicide, probably with the help of an infected friend.371
• Interviews confirmed cyber breach of computer networks and cyber security issues are a
significant concern particularly because they have resulted in several incidents of
information theft. Furthermore in the early 2000s, a computer worm infected the software
of an Iranian uranium enrichment plant, in addition to other industrial sites, affecting
operations of Iranian nuclear centrifuges.372,373 The Department of Defense (DOD)’s
Defense Science Board Task Force considered the potential threat of cyber-sabotage in
their May 2009 assessment of DoD laboratory security, and recommended that an in-
depth study be conducted to determine the potential cyber threat against U.S.
laboratories.374
A summary of the findings drawn from open source literature is presented in Figure 7.5 which highlights
combinations of malicious actors, acts, and consequences of malicious acts based on historical incidents
(in green, red, and red/green circles) and identifies possible combinations based on an evaluation of
malicious actor motivation and capability (blue, blue hatched, or blue diagonal circles). This summary of
findings will be described in detail in the following sections.
371 This case is described in: Seth Carus W, (1998) Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900
Washington, DC, National Defense University.
372 Kushner D, The Real Story of Stuznet. IEEE Spectrum. Accessible at: http://spectrum.ieee.org/telecom/security/the-real-
story-of-stuxnet. Accessed on November 4, 2015.
373
Langner R. To Kill a Centrifuge: A Technical Analysis of What Stuxnet’s Creators Tried to Achieve. Accessible at
http://www.langner.com/en/wp-content/uploads/2013/11/To-kill-a-centrifuge.pdf. Accessed on November 5, 2015.
374 Office of the Under Secretary of Defense For Acquisition, Technology, and Logistics, Defense Science Board, “Report of
the Defense Science Board Task Force on Department of Defense Biological Safety and Security Program,” May 2009, p.
xii, 18-19, 41, <http://www.acq.osd.mil/dsb/reports/ADA499977.pdf>

Figure 7.5. Possible threats based on historical events (green, red and green-red cross-hatched circles) and
malicious actor motivations and capabilities (blue solid, cross-hatched, and patched circles). The green circles
indicate two or more historical events, the red circles indicate one historical event, and the cross-hatched
green-red circles indicate one or two historical events. The blue circles indicate possible threats based on
malicious actor motivation and capability. The patched blue circles indicate planned or failed attempts. The
cross-hatched blue circles indicate limited possibility.
7.2.3 Detailed Descriptions
Detailed descriptions of malicious acts perpetrated by various malicious actors (lone outsider, lone
insider, organized criminals, domestic terrorists and violent extremists, transnational terrorist non-state
groups, and foreign intelligence entities) are included in Appendix V to this report:
Sec. 16.4: Analysis of Malicious Actor Motivations and Capabilities

Sec. 16.5: Detailed Analysis of Historical Incidents
Sec. 16.6: Attacks against laboratories
Sec. 16.7: Biocrimes committed by individuals
Sec. 16.8: Terrorist and extremist events tied to biological warfare (BW)
Sec. 16.9: Designated foreign terrorist organizations and BW
Sec. 16.10: Detailed history of known terrorist BW programs
Sec. 16.11: Other terrorist groups linked in some fashion to BW
Sec. 16.12: Islamic State in Iraq and the Levant (ISIL) group overview
Some types of successful incidents may go undetected, and in general, incidents may be tied to sensitive
law enforcement and intelligence information. Hence, open source reporting alone is unlikely to lead to a
complete list of all relevant historical incidents. Historical patterns can be disrupted, for instance, as a
result of the widespread implementation of new security programs, by the arrest of key group members,
and by the emergence of new malicious actors. Malicious actors may decide, in line with shifts in motives
and capabilities, to change the way they operate and to select new targets. To address the potential
shortcomings of relying solely on historical data, hypothetical events are also considered in light of the
motivations and capabilities of each malicious actor type. That is, when no historical case has been

identified for a particular actor-act pairing, an argument is presented to explain why that pairing is
unrealistic or, on the contrary, for why it cannot be discounted. These cases are called hypothetical.
7.3 Findings: Defense Assessment
An assessment of the overall risk posed by malicious actors necessitates an evaluation of the current
governance structure for biosecurity and related policies, and the implementation of security measures at
research institutions. The historical and operational links between safety and security highlight the need to
include both in this evaluation. 375 In addition, the agents associated with the Deliberative Process –
influenza, SARS-CoV, and MERS-CoV – are subject to different requirements. Highly pathogenic avian
influenza and SARS-CoV are select agents subject to the Biological Select Agents and Toxins
Regulations.376 Although no GoF viruses are currently Tier 1 BSAT, a recent Notice of Proposed Rule-
Making has asked for public input on the upgrade of laboratory-generated, mammalian-transmissible H5
influenza viruses (specifically, those viruses that are contain the HA from the A/Gs/Gd/1/96 lineage and
made transmissible among mammals by respiratory droplets in the laboratory) to the Tier 1 level of
Biological Select Agents and Toxins.377 Low pathogenic influenza and MERS-CoV are not classified as
Biological Select Agents and Toxins. For this reason, included in this analysis are security measures,
whether from governing documents and practices on safety or security, at the non-select agent, select
agent, and Tier 1 select agent levels.
7.3.1 Overview of Security Measures
Table 7.2 below summarizes specific requirements applicable for all laboratories depending on their
biosafety level (second column), additional requirements enforced at laboratories working with Select
Agents and Toxins (third column), and additional requirements enforced at laboratories working with Tier
1 Select Agents (fourth column). The second column constitutes the base level of security, and each
column thereafter lists additional security requirements. The September 2014 institutional DURC
oversight policy applies to select agent and Tier 1 select agent laboratories and to non-select agent
laboratories conducting research with di minimus quantities of botulinum toxin.
375 Biosafety measures mitigate risk of accidental exposure to hazardous biological agents, such as lab acquired infections and
environmental exposure. Biosecurity measures mitigate risk of intentional theft or misuse of biological samples or relevant
sensitive information. U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical
Laboratories – Fifth Edition, December 2009, p. 105, http://www.cdc.gov/biosafety/publications/bmbl5/.
376 42 C.F.R.§73, 9 C.F.R. §121, and 7 C.F.R. §331.
377 Proposed regulation covers laboratory generated, mammalian, respiratory-transmissible influenza viruses containing the
hemagglutinin from the A/Goose/Guangdong/1/96 lineage. Federal Register Volume 80, Number 136, Pages 42079-42084
http://www.gpo.gov/fdsys/pkg/FR-2015-07-16/html/2015-17435.htm.

Table 7.2. Security-specific requirements in general, Select Agent, and Tier 1 Select Agent labs
Select Agent (in addition to Tier 1 Select Agent (in
Topic General
General) addition to Select Agent)
• Appropriate biosafety
training
• Security training, at least
• Dual use research of annually
• Insider threat awareness
Personnel concern training (for training
• Dual use research of concern
Training research involving di • Stricter personnel
training (for research assessed
minimus quantities of reliability reporting
as DURC)
botulinum toxin that is
considered to be DURC)
• Individual security risk
assessment • Pre-access suitability
• DoC and/or DoS permits • Suspicious activity reporting assessment
for pathogen access by
Personnel foreign nationals (if
process • Formal continuous
Reliability needed)1 • Separate criminal background suitability assessment
check (optional) (Behavioral threat
(Reference checks for
new hires; optional) • DoC and/or DoS permits for assessment teams
pathogen access by foreign optional)
nationals (if needed)1
• Self-closing lockable
doors (BSL-2 and up, all
animal)
Separate space from • Three security barriers,
traffic flow, doors locked one monitored
(BSL-3 and up, all • Access control on final
animal) barrier
• Physical security in security
• Separate building or zone, plan • Backup power for
locked doors (BSL-4, access control systems
• Procedures to remove
ABSL-4) • Response time at or
potential malicious actors
Physical • Self-closing doors
• Reporting potential crimes or under 15 minutes, or
(animal) physical barriers
Security access control issues
• Sealed (BSL-3 and up) • Access control management
adequate to hold until
and break-resistant (BSL- responders arrive
4) windows • Inspection of suspicious Restricted off-hours
packages
• (Windows not access even for
recommended for ABSL • Escort visitors approved staff
vivarium; optional) • Procedures for visitors,
(ID badges, access their property, and their
control, “normal” vehicles
working hours; optional)
(Electronic cardkey
access; optional)

Topic General
• Access controls and
training requirements
• Alarmed exits (BSL-4,
ABSL-4)
• Occupational health
monitoring (BSL-4,
ABSL-4, lower levels by
• Intrusion detection
risk assessment
Surveillance systems
• Ventilation alarms (BSL- • No additional requirements
and Monitoring • Occupational health
3 and up, optional below
monitoring
level 4)
• Facility video
surveillance (optional,
generally not monitored)
• Yearly facilities
inspection – biosafety
cabinets, HVAC
• Record number of containers,
storage location, and chain-
of-custody information for
• General inventory and long-term storage
material management • Record animal counts,
Storage, species, location, and final
Inventory, and
process for biological • More stringent reviews,
stocks (optional) disposition logs, and inventory
Accountability
• Record entry/exit in • Access records in logbooks audits (optional)
Processes
logbooks (BSL-4, ABSL- • Access control to inventories
4) • Inventory audits after moving,
PI turnover, or theft/loss
• DoC and/or DoS permits for
pathogen export1
• Triple package agents2
o Labeling requirements • Shipping permits from
for air shipment2 CDC/APHIS required2
o Import permit o Report receipt or
Transfer, • (CDC, USDA), interstate loss/theft/delay to
Shipment, and permit (USDA), potential CDC/APHIS within 48
Chain-of- hours2 • No additional
need for interstate transfer
o Report damage to requirements
Custody permits for imported
Protocols samples (CDC, case-by- CDC/APHIS immediately2
case) 2 • Record transfers
• DoC and/or DoS permits • DoC and/or DoS permits for
for pathogen export (if pathogen export1
needed) 1

Topic General
• External communication
capability (BSL-4,
ABSL-4) • Security response time
Emergency • Annual drills to test at or below 15 minutes,
Response • Emergency access and
emergency and incident or physical barriers
Protocols egress plans (BSL-4,
response plans adequate to hold until
ABSL-4)
responders arrive
• Plans for man-made or
natural disasters (Animal)
General from BMBL3 unless noted. Select Agent and Tier 1 Select Agent from Select Agent regulations4 unless
noted.
1
U.S. Department of Commerce, “Deemed Exports and Fundamental Research for Biological Items”;
15 CFR 734.3-8, “Scope of the Export Administration Regulations”;
15 CFR 744.4-6, “Control Policy: End-User and End-Use Base”;
U.S. Department of Commerce, Commerce Control List, “Category 1 – Special Materials and Related
Equipment, Chemicals, ‘Microorganisms’ and ‘Toxins’”;
22 CFR 121.1(XIV)(b) “The United States Munitions List.”
2
49 CFR 175.134, “Class 6, Division 6.2 – Definitions and exceptions”;
49 CFR 173.196, “Category A infectious substances”;
49 CFR 173.199, “Category B infectious substances”;
49 CFR 172, “Subpart I- Safety and Security Plans”;
9 CFR 122, “Organisms and Vectors,”
42 CFR 71, “Foreign Quarantine”
3
U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories –
Fifth Edition.
4
42 CFR 73, U.S. Government Publishing Office, “Select Agents and Toxins”
9 CFR 121, U.S. Government Publishing Office, “Possession, Use, and Transfer of Select Agents and Toxins”
7.3.2 Detailed Descriptions
A detailed analysis of the requirements, implementation practices, and current gaps in security measures
is provided in Appendix V of this report:
Sec. 16.13: Biosafety and Biosecurity at U.S. Research Laboratories

Sec. 16.14: Laws, Guidance, Policies, Practices, and International Agreements on Biosafety and
Biosecurity
Sec. 16.15: Restriction of Fundamental Research, Dual Use Research of Concern, and NIH
Guidelines for Recombinant DNA
Sec. 16.16: Analysis of Security Measures
Sec. 16.17: Major Challenges and Knowledge Gaps
7.4 Analysis of Offense and Defensive Measures
The biosecurity risk assessment presents a semi-quantitative evaluation about whether deliberate acts
involving GoF influenza, SARS-CoV, or MERS-CoV will result in a local outbreak or pandemic. The
assessment involves: 1) qualitative analysis of plausible threats facing institutions that conduct GoF
influenza virus, SARS-CoV, or MERS-CoV research based on systematic evaluation of historical

incidents, malicious actor motivations and capabilities, and implemented security measures at U.S.
research institutions; and 2) quantitative analysis of the potential for the plausible threats to cause
infections in the local community or broader and of the comparison of possible pandemic consequences
of plausible threats involving GoF influenza virus, SARS-CoV, or MERS-CoV and non-GoF viruses.
Although an actual or attempted biosecurity incident could cause significant damage to research progress,
national preparedness and response efforts, the nation’s economy, or socio-political situation, the
assessment focuses on the consequences to human health (both illness and death) at the individual (i.e.,
laboratory worker, malicious actor, or emergency personnel) and population (i.e., local or global
communities) levels should a pathogen be removed from containment deliberately or accidently.
influenza, SARS, or MERS-CoV virus is available. Therefore, to identify the types of acts that may target
a GoF laboratory, our approach involved examining historical incidents involving life science laboratories
and hospitals, evaluating the motivations and capabilities of malicious actors, and determining if and how
existing security measures affect the likelihood of success of a given malicious act. All of the data
collected on potential threats and biological security governance were used to assess the plausible threats
facing laboratories that study or store GoF virus(s).378 For the purpose of this analysis, “plausible threats”
are defined as the most probable events that could lead to a loss of containment from a biosecurity
incident. Therefore, the analysis focused on the plausible threats assessed within the current context of
laboratory security and their potential to lead to localized or widespread infections.
The malicious acts that present the greatest risk to human health are assessed sequentially, starting with
malicious actors. The most plausible actors are further evaluated by the most probable and consistent
malicious acts they may commit. Finally, the most likely immediate consequence of a probable and
consistent act that has been committed is evaluated. At each step, probable threats are evaluated within
the context of current security measures at U.S. high containment research laboratories. The final result is
the most plausible threats based on evaluation of historical data, consistency with malicious actor
motivation and capability, and likelihood within the current security environment at high containment
research laboratories in the United States.
The potential of plausible malicious acts to cause global pandemic was assessed using the biosafety risk
assessment models. By leveraging the biosafety risk assessments to analyze biosecurity risk, the
important input parameters become the number of initial infections and response time after an incident
(including emergency response and/or public health response). Deliberate and accidental risks result in
vary similar outcomes and this approach allows for comparisons to be made between biosafety and
biosecurity risks that cause similar human health outcomes.
7.4.1 Qualitative Assessment of Plausible Threats
7.4.1.1 Malicious Actor
Analysis of malicious actor intent falls into two categories: 1) intent to target U.S. research institutions to
acquire GoF viruses for use as weapons; and 2) intent to harm U.S. research institutions and/or laboratory
workers, but not through the weaponization of pathogens stored or studied in research laboratories. Table
7.3 summarizes the likelihood that the malicious actors considered have the intent, capability and ability
to access laboratories with respect to each of these categories. The analysis is based on an evaluation of
historical cases and malicious actor motivations and capabilities described in Sec. 7.4 and Appendix V
378
Noncompliance with security regulations might increase biosecurity risk intentionally or unintentionally. However, because
no repository of tested best practices exist, some requirements may not be easily implemented at research institutions given
building design, institutional policies, and local and state laws.

Sec 16.2-16.9 and the safety and security measures included in Sec. 7.5 and Appendix V Sec 16.10-16.11.
Although this assessment is grounded in historical incidents, incorporation of malicious actor motivations
and capabilities ensured that plausible incidents that have not previously occurred would be considered.
Of greatest relevance to the discussion about actors is their definition: outsiders are not authorized to
access high containment research laboratories in which GoF research with influenza, SARS-CoV, or
MERS-CoV are conducted, while insiders are authorized to access such laboratories by definition. The
approach taken in this analysis can be applied to biosecurity risk assessments of research involving other
pathogens.
Table 7.3. Malicious Actor Intent, Capability, and Opportunity

Deliberate Acts that Use of Deliberate Act Resulting in Accidental
Pathogens as Weapons Release of Viruses
Intent Intent to Capability
to Capability Ability to Carry to Carry Ability to
Acquire to Acquire Access Out Out Access
Virus to Virus Laboratory Malicious Malicious Laboratory
Use Act Act
Foreign Intelligence
Agencies
Transnational
Terrorists, non-state
actors
Domestic Terrorists
and
Extremists
Organized Criminals
Only to outside
Lone Outsiders
of building
Lone Insiders
Black indicates consistency of intent, capable, or able to access high containment, research laboratories with
known malicious actor motivations and capabilities, and historical incidents.
Dark Grey indicates possible intent, capable, or able to access high containment, research laboratories with
Grey indicates inconsistency of intent, capable, or able to access high containment, research laboratories with
7.4.1.1.1 Deliberate Act at U.S. Research Laboratory to Use Pathogen as Weapon
The most likely actors with intent to target U.S. research institutions to acquire GoF influenza virus,
SARS-CoV, or MERS-CoV for use as weapons are transnational terrorists and lone insiders.
• Although foreign intelligence agencies may want to acquire viruses, their purpose for doing so is
likely for intelligence, scientific advancement in their home countries, or commercial benefit.
Although the possibility that foreign intelligence agencies may want to acquire GoF virus to

incorporate into their offensive biological weapons cannot be ruled out, the likelihood of this
intent is low. Approximately 172 countries are party to the Biological Weapons Convention,
which bans development and stockpiling of biological weapons, and all nations are required to
abide by United National Security Council Resolution 1540, which requires countries to
implement and enforce measures preventing proliferation of biological weapons within their
borders. These international obligations decrease (but does not eliminate) the likelihood that
Nations acquiring a virus for an offensive biological weapons program.
• Translational terrorists, specifically al Qaeda, continue to express interest in acquiring pathogens

for use as weapons. In addition, al Qaeda and ISIL have recruiting efforts that target individuals
with technical skillsets to join their causes and undertake malicious acts within their means.
However, no available information suggests that these groups have recruited scientists in the U.S.
or that other transnational terrorist groups have interest in biological weapons. For these reasons,
the intent of transnational terrorists is described as possible.
• In our analysis, individuals who self-radicalize and plan to or carry out a malicious act in the
absence of a formal affiliation with a terrorist organization are considered Lone Outsiders or Lone
Insiders. This distinction is made because the level of resources, support and success afforded a
member of a group compared to an individual acting alone is different, all of which will be
described in the analysis of capabilities, access, and likelihood of malicious acts.
• Historically, members of domestic terrorist groups, but not animal rights extremist or eco-radical
groups, have sought to acquire bacteria from culture collections. Similarly, recent policy debates
about synthetic genomics have raised concerns that individuals, some of whom may be members
of these groups, may seek to acquire viral DNA from DNA synthesis companies to recreate
viruses. However, no available examples exist describing cases where domestic terrorist groups
have sought to steal or have successfully stolen viruses from high containment laboratories in the
United States, suggesting the intent to do so is inconsistent with their motivations. However, one
example exists describing a domestic terrorist group that used an agent against the public to sway
a local election (Rajneeshee Cult).
• Organized criminals and lone outsiders are not likely to be interested in stealing virus from a high
containment laboratory based. on the lack of open source examples of such incidents and our
understanding of the motivations of organized criminals.379
• Several historical cases involving lone insider theft and use of bacteria to harm others were
identified in open source literature during our study. These cases have involved disgruntled,
dissatisfied, disturbed, or radicalized insiders who remove a pathogen from a laboratory to infect
co-workers, spouses, or family members. Extrapolating these cases to GoF viruses, the possibility
that a lone insider, with malicious intent, may acquire virus from a research laboratory to use as a
weapon is high.
When divorcing intent from capability, the most likely malicious actors to have the capability to acquire
GoF influenza virus, SARS-CoV, or MERS-CoV from U.S. research laboratories are foreign intelligence
entities, an insider acting alone or in concert with any category of malicious actors, and lone insiders.
379 Historical examples of lone outsiders acquiring bacteria from culture collections do exist. But, to our knowledge, culture
collections do not have GoF influenza, SARS-CoV, or MERS-CoV viruses.

• Many foreign intelligence agencies have the resources and levels of expertise to suggest they
have a high level of capability. In addition, these agencies are known to elicit information from
insiders as part of their typical tradecraft.
• Transnational terrorists, domestic terrorists, and domestic extremists themselves are unlikely to
have the requisite capability to illicitly acquire a virus from a high containment laboratory in the
United States. However, documented historical cases describing insider recruitment suggest that
insider-assisted capability is likely.
• Organized criminals and lone outsiders are unlikely to have the capability to steal virus from a
high containment laboratory, particularly since no open source examples exist. However despite
the lack of motivation, the possibility that organized criminal groups force insiders to assist in
acquiring virus cannot be ruled out.
• Lone insiders present the highest risk when considering capability because they have the
knowledge, skills, and access necessary to acquire and/or manipulate the viruses.
When considering an actor’s ability to access a GoF influenza virus, SARS-CoV, or MERS-CoV in a
high containment research laboratory in the United States, the most likely malicious actors to have access
are foreign intelligence entities, an insider acting alone or in concert with any category of malicious
actors, and lone insiders.
• Research institutions supporting animal research and high containment research have access
controls in place to protect against unauthorized access to the laboratories. These controls can
take the form of guards; electronic, biometric, or mechanical intrusion prevention/detection
systems; and/or some combination of these measures. Based on these access controls, along with
periodic monitoring of access to laboratories, the likelihood that any outsider, who is not working
with an insider, could gain access to the high containment, research laboratory to acquire GoF
virus is extremely low. That said, these security measures are only as good as the community who
observe them (i.e., noncompliance with security regulations might increase insider-assisted
biosecurity risks),
• Foreign intelligence agencies are known to elicit information and materials from insiders as part
of their typical tradecraft, suggesting the possibility of gaining access to GoF viruses through
indirect means. In addition, these agencies may have personnel who are authorized entrance into
the laboratories and have direct access to GoF viruses (e.g., operative or elicited individual).
Professionals likely would not be identified by currently implemented personnel reliability
measures at research institutions or deterred by access control measures.
• Acting alone, translational terrorists are not likely to have any opportunities to acquire virus in
high containment research laboratories in the United States. However, these organizations are
known to recruit individuals to their causes, suggesting that they could acquire virus with the help
of an insider (e.g. through elicitation, subversion, or recruitment). Current insider threat training
and personnel reliability measures that allow for periodic behavioral assessment, specifically
implemented for Tier 1 BSAT, and non-punitive reporting of changes in co-worker behavior (any
level of research) could alert institutional officials to possible insider radicalization. However,
colleagues may not recognize such changes or may be in denial that such changes are taking place
in a friend or colleague, limiting the effectiveness of personnel security measures.

• Domestic terrorists, domestic extremists, organized criminals, and lone outsiders are not likely to
have access to virus stored in high containment, research laboratories in the United States.
Despite the lack of access, the possibility that domestic terrorists, domestic extremists, and
organized criminals could acquire a virus with the help of an insider cannot be ruled out.
Historical examples of domestic extremist groups gaining access to lower containment
laboratories and eliciting information about facilities or attempting to get into animal facilities
exist. However, the likelihood that such elicitation and access attempts would translate to
acquisition of GoF virus is low in light of current access controls for high containment research
laboratories, particularly BSAT laboratories, and personnel reliability measures for Tier 1 BSAT
laboratories. Increased insider vigilance and non-punitive reporting would further decrease the
likelihood of insider-assisted acquisition,
• Lone insiders are extremely likely to have opportunities to acquire virus from high containment
research laboratories because they have authorized access to these laboratories. The monthly
inventory checks on stored pathogens would not necessarily deter insiders (both lone insiders and
insiders assisting groups) from removing virus from the laboratory. In addition, the identification
of missing virus may be impossible if some virus is removed from a vial that remains in the
freezer. In addition, inventory checks would not identify removal of virus from experimental
samples, and
• The possibility that an actor could steal pathogen during transportation appears to be low because
GoF viruses apparently are not shipped.380 Even if a virus was shipped, specific information about
shipping dates, trucks, and vendors are not accessible to outsiders.
When evaluating intent, capability, and opportunity together, the most likely malicious actor to target a
U.S. research laboratory to acquire a pathogen for use as a weapon is an insider, either working alone or
in coordination with a group, likely a transnational terrorist group.381
7.4.1.1.2 Deliberate Act at U.S. Research Laboratory Resulting in Accidental Release of Viruses
Deliberate acts directed towards institutions and people, but not conducted for explicit acquisition of virus
for use as a weapon, could target: 1) the research laboratory in which GoF research with influenza virus,
SARS-CoV, or MERS-CoV is being conducted; 2) space outside the laboratory but inside the building
which houses the laboratory; or 3) the area outside the facility in which the laboratory is housed. Such
targeting could cause accidental release of virus from the laboratory. Specific acts associated with such
targets could include armed assault, arson, bombing, vandalism and sabotage of facilities, tampering with
experiments, and theft of materials, equipment, and animals.
The most likely malicious actors with the intent to carry out such acts include domestic terrorists and
extremist groups, and lone insiders.
• The possibility that a foreign intelligence agency would carry about a deliberate act on a U.S.
research laboratory is low, especially since an attack could be construed as an act of war. This is
particularly true for overt attacks, such as bombing, armed assault, or vandalism. However, the
possibility that a foreign intelligence agency could accidently release a GoF virus through theft of
materials or equipment cannot be ruled out.
380
Scientists use reverse genetics to make GoF viruses instead of shipping them according to the scientists who were
interviewed.
381 An insider also is the most likely actor to acquire a virus for non-weapons purposes, such as personal or monetary benefit or
assisting foreign intelligence agencies.

• No open source information exists indicating an intent by transnational terrorists to carry out a
deliberate, malicious act on high containment research laboratories in the United States.
However, the possibility that transnational terrorists may want to bomb a building cannot be ruled
out because of the high prevalence of such tactics by several of these groups.
• No open source information exists indicating an intent by organized criminal groups to carry out
a malicious act on or in high containment research laboratories in the United States. Organized
criminals are driven by financial gain suggesting that the possibility that a criminal organization
might seek to sell laboratory equipment for profit cannot be ruled out. However, the relative
availability of common life science equipment for online purchase decreases the likelihood that
an organized criminal organization will steal from a high containment laboratory in the United
States.
• Several historical cases involve deliberate acts caused by domestic terrorists and extremists who
have vandalized buildings, tampered with experiments in lower containment laboratories,
released research animals into the wild or their own homes, or detonated bombs near buildings.
The frequency with which these groups attack research institutions for ideological purposes
indicates a high likelihood that they could carry out more such deliberate acts. However,
additional security measures put in place under various regulations make such acts much more
difficult to plan and carry out.
• Historical cases involving the use of bombs or armed assault in public areas and at hospitals by
lone outsiders exist in open source literature. However, no open source information was identified
about the targeting of U.S. research laboratories by lone outsiders. Despite the lack of motivation,
the possibility that a lone outsider would detonate a bomb or carry out an armed assault outside or
in a research building cannot be ruled out. However, the motivation for such an attack is not clear.
• Historical examples of lone insiders tampering with experiments for personnel benefit or theft of
virus for commercial benefit suggest the presence of clear motivations for lone insiders to carry
out deliberate acts, some of which could result in accidental release of virus. Consequently, the
likelihood of such acts is high.
When considering only capability, the most likely malicious actors to carry out deliberate acts that might
result in accidental release of virus are foreign intelligence agencies, transnational terrorists, domestic
terrorists and extremists, and lone insiders.
• Foreign intelligence agencies are expected to have the resources, tools, and expertise needed to
carry out deliberate acts that could result in accidental release of GoF virus. In addition, these
agencies are known to elicit information and materials from insiders as part of their typical work.
• No open source information about transnational terrorists targeting U.S. research laboratories
exists. However, the 2001 attacks and the violence carried out by individuals who may have been
radicalized by transnational terrorists suggests their capability to carry out armed assault, arson,
and bombing within the United States. Whether surveillance and monitoring of building
perimeters would deter or prevent a transnational terrorist or sympathizer from carrying out an
attack using these tactics is unclear.
• Domestic terrorist and extremist groups have bombed hospitals, released animals from research
laboratories, vandalized research laboratories and equipment, and tampered with experiments.

Their proven ability to damage the building exteriors, damage low containment laboratories, steal
animals, and tamper with experiments suggests they are capable of damaging buildings in which
GoF research is being conducted. However, domestic terrorist and extremist groups are not likely
to carry out deliberate acts inside a high containment research laboratory without the assistance of
an insider with access. Increased security, including surveillance and monitoring of building
perimeters and animal facilities, and increased arrests has decreased deliberate, violent acts
involving animal rights extremists. Whether this extrapolates to other domestic terrorist or
extremist group is unclear.
• No open source information exists about the capability of organized criminals to damage
biological research facilities in the United States deliberately. Despite this unknown capability,
organized criminals could use armed assault to gain access to the facility, but the exact purpose of
doing so is unclear.
• Lone outsiders have detonated bombs in public areas and at certain clinics suggesting their
potential capability to damage buildings in which GoF research is being conducted. Surveillance
of building perimeters may deter lone outsiders from carrying out such acts. However, the
increasing number of active shooter incidents at U.S. facilities and educational institutions
suggests that such actors are not deterred by surveillance and other similar measures.
• Lone insiders are expected to have the knowledge and skills to deliberately compromise or
tamper with equipment and experiments. Furthermore, historical cases involving lone insiders
who tamper with co-workers’ experiments are described in open source literature. These cases
and the presumed knowledge and skills of lone insiders suggest that these actors likely have the
requisite capabilities to carry out deliberate acts in a high containment research laboratory. Non-
punitive peer reporting of unusual incidents or repeated experimental findings, damaged
equipment and facilities, and behavioral changes or unusual behavior of individuals with
authorized access to high containment, research laboratories are the only measures that exist to
prevent or mitigate a deliberate act carried out by an insider with trusted access.
When analyzing an actor’s ability to access a high containment, research laboratory, the most likely
malicious actors to carry out deliberate acts that could result in accidental release of virus are foreign
intelligence agencies, transnational terrorists, domestic terrorist and extremist groups, and lone insiders.
• Research institutions supporting animal research and high containment research have access
controls in place to protect laboratories. These controls can take the form of guards; electronic,
biometric, or mechanical intrusion prevention systems; and/or some combination of these
measures. Based on these access controls and periodic monitoring of access to laboratories, the
likelihood that any outsider, who is not working with an insider, would gain access to the high
containment, research laboratory to tamper with experiments involving GoF viruses or their wild-
type counterparts is low.
• Foreign intelligence agencies are known to elicit information from insiders as part of their typical
tradecraft, suggesting the ability to achieve indirect access to laboratory materials. In addition,
these agencies may have personnel who can gain authorized entry to the laboratories (i.e.,
insertion of an operative) for direct access to high containment laboratories. Professionals likely
would not be identified by currently implemented personnel reliability measures at research
institutions or deterred by access control measures. Foreign intelligence agencies also may have
the resources to access laboratories remotely by hacking into laboratory computer systems.

Finally, foreign intelligence agencies may have the resources to detonate a bomb or carry out an
armed assault, but as previously stated, these acts could be construed as an act of war.
• Transnational terrorists may have access to the exterior perimeter of the building in which a
laboratory is located and potentially to the research laboratory itself, if access control measures
are insufficient. However in general, such actors would not have access to the high containment
laboratory itself unless assisted by an insider. Current personnel reliability measures that allow
for periodic behavioral assessment, specifically implemented for Tier 1 BSAT, and non-punitive
reporting of changes in co-worker behavior (any level of research) could alert institutional
officials to possible radicalization of an insider. That said, colleagues may not recognize such
changes or may be in denial that such changes are taking place in a friend or colleague.
• Historically, domestic terrorist and extremist groups, such as animal rights extremists, have
recruited, elicited information from, and subverted insiders to gain access to animal facilities. In
addition, other groups have elicited information about clinics as they prepared to bomb buildings
based on historical examples. Domestic terrorist and extremist groups are likely to access the
perimeters of buildings and low containment research laboratories, but not likely to access high
containment research laboratories without the assistance of an insider.
• The likelihood that criminal organizations and lone outsiders would have access to high
containment, research laboratories is low. However, the possibility that an insider could assist a
criminal organization in carrying out a deliberate act in a high containment laboratory cannot be
ruled out, though the exact purpose behind such an act is not clear.
• Lone insiders are extremely likely to have opportunities to tamper with experiments, release
animals, compromise equipment, detonate bombs, and carry out armed assault in high
containment, research laboratories because they have authorized access to these laboratories.
7.4.1.1.3 Malicious Actor Conclusion
When evaluating the intent, capability, and the ability to access laboratories in which GoF research with
influenza virus, SARS-CoV, or MERS-CoV, the most likely malicious actors to target a U.S. research
laboratory to carry out a deliberate act to the building perimeter are domestic terrorists and extremists,
transnational terrorists, lone outsiders, and lone insiders. Although no open source information indicates
whether these malicious actors are motivated to damage buildings in which GoF viruses are stored or
studied, historical examples of attacks involving other types of buildings do exist.
When looking at all three components together for deliberate acts carried out inside a high containment
research laboratory in which GoF research with influenza virus, SARS-CoV, or MERS-CoV is
conducted, the most likely malicious actor is an insider, working alone or in coordination with a group,
particularly domestic terrorist or extremist groups.
7.4.1.2 Malicious Acts and Likelihood of Escape of GoF Virus
The likelihood of success of malicious acts and resulting virus escape are based on the degree of access to
a high containment research laboratory. These laboratories (i.e., biosafety levels 3 and 4) have a variety of
security measures in place to prevent unauthorized access by individuals who are not approved to work
and/or do not demonstrate competency and proficiency in working safely and competently in the
laboratory. In addition, researchers working with BSAT are subject to review by the Security Risk
Assessment (Appendix V Sec 16.11.2) and those working with Tier 1 BSAT must undergo periodic
screening assessments. Based on these physical and personnel security measures, the analysis of

malicious acts is divided into: 1) acts that can be carried out by only insiders; and 2) acts that outsiders
can be carried out without insider assistance. The analysis draws upon historical cases to evaluate the
likelihood that malicious acts would be undertaken successfully and to focus on those acts that likely
could cause a breach leading to escape of a GoF influenza virus, SARS-CoV, or MERS-CoV. Table 7.4
summarizes the likelihood of an outsider or insider to successfully carry out a particular malicious act and
the likelihood that such an act could lead to GoF virus escape.
Table. 7.4. Malicious Acts Undertaken and Likelihood of Success

Leads to GoF Virus
Outsider Insider
Escape
Armed Assault
Depends of size,
Bomb type, and location of
a bomb blast
Arson
Physical Entry N/A Unlikely by itself
Cyber Breach
Infect Co-
Theft of Virus Workers
Infect Public
Theft of Animals
Theft of Materials, Equipment, or
Information
Sabotage
Elicitation of Information N/A
Subversion of Employees N/A Unlikely by itself
Insertion of Operative N/A Unlikely by itself
Reckless Act Depends on the act
Deliberate Self-Infection
Armed Assault
The increase in active shooter incidents in the U.S. suggests that an armed assault in a high containment
research laboratory may be possible at some level. Outsiders could carry out an armed assault outside the
building in which GoF research is conducted. Insiders possibly could carry out an armed assault inside a
research building and high containment research laboratory.
Current personnel security measures requiring periodic assessment of personnel and non-punitive
reporting of behavioral changes in personnel could provide an opportunity for institutional officials to
identify potential insider threats before acts are conducted. These measures are required for Tier 1 BSAT
laboratories. In addition, institutions conduct emergency response exercises and many universities have
threat assessment teams to evaluate the threats on campus and identify prevention strategies. Physical
security measures, including physical barriers, access controls, and surveillance measures, work to

prevent armed outsiders from gaining access to high containment, research laboratories. However, no
physical security measures are in place that would prevent authorized insiders from taking guns into high
containment research laboratories.
An armed assault leading to escape of a GoF virus is unlikely even if the assault is carried out
successfully. Exposure to GoF virus through an open wound is unlikely to cause infection. However,
active shooters inside a laboratory might lead to viral escape through accidental aerosolization of virus in
experimental samples (i.e., exposing the shooter to aerosolized GoF virus or contaminating street clothing
with fomites). If emergency personnel also enter the laboratory, they may be exposed to aerosolized virus
or fomites if not wearing proper protection.
Bomb or Arson
Several historical cases exist of malicious actors detonating bombs in public areas, outside buildings, or at
clinics, or setting fires to research buildings. Although outsiders could detonate bombs outside of
buildings or areas accessible to the public, they do not have access to high containment research
laboratories unless assisted by an insider. Insiders potentially could detonate bombs inside research
buildings or high containment research laboratories.
As with armed assault, current Tier 1 personnel security measures could provide opportunities to prevent
insiders from successfully detonating a bomb or setting a fire in a research laboratory and building.
Institutions conduct emergency response exercises and universities have threat assessment teams to
evaluate the threats on campus and identify prevention strategies. Although physical security measures
help prevent outsiders from gaining access to high containment research laboratories, these measures
would not prevent authorized insiders from detonating bombs or setting fires in high containment research
laboratories.
The size, type, and location of a bomb blast may lead to escape of GoF virus from experimental samples.
If the size of the blast is sufficiently large to rattle the building infrastructure (to a similar degree as an
earthquake), GoF virus might aerosolize from spilled experimental samples leading to possible loss of
containment. Similarly, a blast that occurs at the entrance or inside of a high containment, research
laboratory might result in aerosolization of GoF virus and compromise the negative pressure of the
laboratory. However, arson is unlikely to lead to escape of a GoF virus even if carried out successfully
because institutions have established procedures and response measures for fires and viruses are sensitive
to high temperatures.
7.4.1.2.1 Physical Entry
By definition, insiders have authorized access to high containment research laboratories. Consequently,
this type of act does not apply to insiders. Outsiders are unlikely to gain physical access to high
containment, research laboratories without assistance from an insider.
Physical security measures employed at high containment research laboratories and animal facilities,
including physical barriers, access controls, and surveillance measures, help prevent outsiders from
gaining access to the laboratory itself.
Physical entry alone does not lead to escape of GoF virus from containment.

Cyber Breach
Over the past decade, a growing number of malicious actors, from nation-states to individuals, have
hacked into computer systems in the pharmaceutical, health care, insurance, national security, and
commercial organizations. Furthermore in 2010, a computer worm that infected the software of an Iranian
uranium enrichment plant, in addition to other industrial sites, affecting operations of Iranian nuclear
centrifuges382,383 suggesting this attack approach should not be ruled out.
Outsiders or insiders with the requisite expertise could hack into the computer systems of research
institutions. Other than firewalls, anti-virus software, and standard cyber security measures, no specific
measure is required to protect information and infrastructure systems from cyber breaches. The exception
is Biological Select Agents and Toxins laboratories, which are required to have information security in
place to prevent cyber breaches.384,385,386,387,388,389
The likelihood that a cyber breach would lead to escape of GoF virus is moderate. However, breaches in
infrastructure systems, such electronic controls for air-handling, could lead to escape pf a GoF virus from
containment. That said, air-gapped systems (i.e., those systems that are not connected to the internet) are
much less likely to lead to escape of GoF virus. Based on our interviews, systems that control laboratory
operations, air filtration, and decontamination are not connected to the open internet, but this does not
necessarily mean that the systems are immune to attack.
Theft of GoF Influenza virus, SARS-CoV, or MERS-CoV
Theft of GoF virus could occur in two ways: 1) by stealing it from a high containment, research
laboratory; and 2) theft or diversion during transportation. Outsiders acting without the assistance of an
insider likely are not able to steal GoF virus from high containment laboratories because of the various
access control measures in place to prevent unauthorized access into these laboratories. The likelihood of
an outsider stealing GoF virus during transportation is low primarily because knowing the transfer date,
exact truck carrying the virus, and transportation line used would be extremely difficult without assistance
by a knowledgeable insider.
The likelihood of an insider successfully stealing a GoF virus from a laboratory is high because by
definition, such an individual has authorized access to high containment research laboratories.
Furthermore, insiders have used pathogens against co-workers, family members, and members of the
public in the past. Current personnel security measures requiring periodic assessment and non-punitive
reporting mechanisms could identify behavioral changes in personnel before an act is committed.
Inventory measures, could help identify discrepancies in stored GoF virus. However, theft of virus could
occur in-between the regular inventory reviews, could be missed if virus is removed from vials, or could
be via theft of experimental, infectious samples.
Theft of GoF virus leads to escape of virus from containment by definition, but such an act does not
presume that the stolen virus could be used as a weapon. A significant amount of processing, including
382 Kushner D, The Real Story of Stuznet. IEEE Spectrum. Accessible at: http://spectrum.ieee.org/telecom/security/the-real-
story-of-stuxnet. Accessed on November 4, 2015.
383
Langner R. To Kill a Centrifuge: A Technical Analysis of What Stuxnet’s Creators Tried to Achieve. Accessible at
http://www.langner.com/en/wp-content/uploads/2013/11/To-kill-a-centrifuge.pdf. Accessed on November 5, 2015.
384 42 C.F.R. §73.11(c)(1).
385 42 C.F.R. §73.11(c)(9).
386 9 C.F.R. §121.11(c)(1).
387 9 C.F.R. §121.11(c)(9).
388 7 C.F.R. §331.11(c)(1).
389 7 C.F.R. §331.11(c)(9).

growth of the virus from a frozen stock, may have to be carried out to make the virus usable and/or
disseminable. However, theft of experimental samples that contain GoF viruses might be used as is.
Theft of Animals
Several historical examples exist of domestic animal rights extremists removing animals from low
containment research laboratories to take home as pets or release into the wild. However, no examples
exist for high containment research laboratories or animal housing facilities, likely because of increased
security and access controls of both facilities. The physical security and perimeter surveillance measures
of facilities where research animals are present have increased to counter deliberate acts carried out by
animal rights extremists. Current personnel security measures involving periodic assessment (as in Tier 1
BSAT) and non-punitive reporting mechanisms might identify insiders who have been elicited, recruited,
or subverted by outsiders or decided to carry out a malicious act on his/her own. Vigilance by other
laboratory workers could decrease the likelihood that animals go missing.
Theft of infected animals would likely lead to escape of GoF virus.
Theft of materials, equipment, and information
Despite numerous historical cases involving deliberate acts carried out by domestic terrorists and
extremists, none have involved high containment research laboratories. The historical cases evaluated
involved assistance from insiders to provide information, enable access into laboratories, or carry out
actual deliberate acts, such as theft of animals or vandalism of equipment. Physical barriers and access
control measures prevent outsiders from gaining access to high containment research laboratories.
Therefore, outsiders acting without insider-assistance are unlikely able to steal laboratory materials,
equipment, and information contained in high containment research laboratories. Because insiders are
authorized to access high containment research laboratories, the likelihood that they could steal materials,
information, or equipment is high.
Current personnel security measures involving periodic assessment (as in Tier 1 BSAT) and non-punitive
reporting mechanisms might identify insiders who have been elicited, recruited, or subverted by outsiders
or decided to carry out a malicious act on his/her own. Vigilance by other laboratory workers could
decrease the likelihood that equipment, materials, or information (e.g., laboratory notebooks or inventory
logs) go missing.
Theft of contaminated equipment or materials might lead to escape of GoF virus. Theft of information
would not lead to escape of GoF virus.
Sabotage
The likelihood that outsiders will tamper with equipment or experiments in high containment is low if not
assisted by an insider. The likelihood that an outsider will tamper with the laboratory itself (including the
HEPA filtration system and waste management system) is low unless such an individual has assistance
from a knowledgeable insider. An insider with access to experiments and equipment could tamper with
them. However, not all insiders have access to laboratory operating systems, reducing the likelihood of
such acts.
Current physical barriers and access controls help prevent outsiders from gaining access to high
containment, research laboratories and their primary operating systems. Personnel security measures help
identify insiders who might carry out acts of sabotage within the laboratory, against a facility, or in the

laboratory operating system. However, these measures would not necessarily enable detection of insiders
in a non-BSAT high containment research laboratory.
Sabotage of experiments, equipment, or laboratory operating systems might lead to escape of GoF virus.
For example, tampering with laboratory materials could lead to ineffective decontamination of samples,
which, if undetected, could result in accidental exposure of laboratory workers who don’t realize the viral
samples are still infectious. Other examples include removal of HEPA filters from the air flow system,
which would prevent proper filtration of the laboratory air, or tampering with a centrifuge rotor, which
could result in an imbalance during spins causing the contents to rupture and exposing laboratory workers
to the infectious samples.
Reckless Act
Reckless acts include mixing of infected animals with uninfected animals to deliberately tamper with
experiments. Several historical cases involving animal rights extremists suggest that these acts can be
carried out in low containment research laboratories. However, the increase in physical barriers, access
control measures, surveillance of animal facilities, and arrests have deterred such groups from carrying
out these acts.
These types of reckless acts are highly implausible for GoF virus research, because infected and
uninfected experimental animals are kept in high containment research laboratories, which often are in
different locations than facilities housing uninfected animals that are not part of ongoing research.
Consequently, the likelihood that an outsider could carry out such acts is low unless assisted by an
insider. The likelihood that insiders who have been elicited, recruited, or sabotaged by an outside group
could remove animals from high containment, research facilities is high. However, because animals
involved in active experiments are separated physically from animals not involved in experiments
suggests that mixing of infected and uninfected animals in lower containment is not as likely, though not
impossible.
Physical barriers, access control measures, and surveillance of animal facilities and BSAT facilities help
prevent outsiders from entering high containment, research laboratories unassisted. Current personnel
reliability measures, including periodic assessment and non-punitive reporting mechanisms, help
institutional officials to detect changes in behavior in personnel. However, these personnel security
measures are required only for Tier 1 BSAT laboratories; some non-Tier 1 BSAT laboratories implement
these measures on their own or as part of their institution’s Tier 1 BSAT program, if applicable.
Reckless acts, such as removal of experimental animals, could lead to escape of a GoF virus via the
infected animal. Mixing of infected and uninfected animals could lead to escape of a GoF viruses if the
uninfected animals are in low containment and in contact with people. Deliberate infection of oneself, co-
worker, friend, or family member leads to escape of a GoF virus.
Deliberate Self-Infection
Two historical cases of deliberate self-infection exist; however, these cases do not involve self-infection
with a virus taken from a research laboratory. Acts involving deliberate self-infection require an actor to
obtain the GoF virus either from a high containment research laboratory. Outsiders acting without the
assistance of an insider are not likely to obtain a GoF virus from high containment laboratories because of
the various physical barriers and access controls in place at such laboratories. The likelihood of an
outsider obtaining GoF virus during transportation is low primarily because knowing the transfer date,
exact truck carrying the virus, and transportation line used is impossible without assistance by a
knowledgeable insider.

The likelihood of an insider obtaining a GoF pathogen from the laboratory for use in self-infection is high
because (s)he has authorized access to high containment research laboratories. Current personnel security
measures requiring periodic assessment and non-punitive reporting mechanisms could identify behavioral
changes in personnel before an act is committed. The benefit of inventory measures for acquisition of
virus for self-infection is unclear. The insider likely would use experimental samples, which are not part
of current long-term storage measures.
Deliberate self-infection of insiders would lead to escape of GoF virus from containment.
7.4.1.2.2 Malicious Act Conclusion
The most likely malicious acts that could lead to escape of a GoF virus from high containment research
laboratories are: theft of GoF virus or contaminated equipment; tampering with experiments or laboratory
operating systems; removal of infected animals; and deliberate infection of oneself, friend, family
member, or co-worker. All of these acts involve insiders, who are either acting alone or in coordination
with a group, such as domestic terrorist and extremist group. Whether theft of GoF virus leads to
exposure and infection by laboratory workers or members of the public depends on the virus’ form (either
from frozen vials or experimental samples) and the skills and resources of the malicious actor to
effectively grow and deliver the virus.
A possible malicious act that is less likely to lead to escape of a GoF virus is a bomb. The size and
location of a bomb determines whether its detonation could lead to escape of GoF virus in experimental
samples. Outsiders and insiders could detonate a bomb, though in different locations, i.e., outside the
building or near a high containment, research laboratory.
7.4.1.3 Type of Breach Leading to GoF Virus Escape
The likelihood of virus escape and human infection caused by malicious acts are summarized in Table
7.5.

Table 7.5. Type of Breach Leading to Virus Escape
Malicious Act GoF Virus Escape Human
Infection
Theft of Animals
Release of Infected Animals
Sabotage
from and within Laboratories
Reckless Act
Release of Infected Animals Theft of Animals
in the Environment Reckless Act
Loss of Containment
Cross-Contamination of Sabotage
Laboratory Animals Reckless Act
Armed Assault
Exposure of Laboratory Bomb
Workers
Sabotage
(Could Include Emergency
Personnel accessing the Reckless Act
Laboratory) Deliberate Self-
Infection
Removal of GoF Virus from
Theft of GoF Virus
the Laboratory
Infection of Wild or Domestic Theft of GoF Virus
Release of GoF Virus
Deliberate Outdoor
Animals Theft of Animals

Infection of Laboratory
Theft of GoF Virus
Workers
Theft of GoF Virus
Infection of the Public Theft of Materials
Theft of Equipment
Release of Infected Animals from and within Laboratories.
The design of and security measures associated with high containment research laboratories help prevent
unassisted escape of animals from the laboratory. However, if an insider intentionally releases laboratory
animals outside of high containment research laboratories, the likelihood that animals will escape the
building varies based on the number of released animals, the method of release, their ease of capture by
researchers, and the design features of the facility, all of which either limits or permits animal escape.
Furthermore, the number of people that the animal might encounter as it wanders around in the building
affects the level of exposure these individuals have to GoF virus from infected animals. Because of this
variability, the likelihood of GoF virus escape and human infection resulting from theft of infected
animals from and within laboratories is moderate and depends on a variety of factors.
Release of Infected Animals in the Environment
Theft of animals would result deliberate release of infected animals into the environment, whether in the
wild or someone’s home (as a pet), and hence, would be considered as a GoF virus escape. Furthermore,
the close proximity of the infected animal to the actor who releases the animal suggests that at least one
human (the malicious actor) would be exposed to the GoF virus and could be infected.

Cross-Contamination of Laboratory Animals
Sabotage of experiments, including the deliberate mixing of infected and uninfected animals within high
containment research laboratories, neither increases the likelihood of GoF virus escape, nor increases the
likelihood of human infection. However, the deliberate mixing of infected and uninfected animals in
lower containment research environments (i.e., sabotage) might expose researchers not protected against
H5 influenza, SARS-CoV, or MERS-CoV to GoF virus, and cause GoF virus escape if the virus gets on
street clothing. The likelihood that this exposure could result in human infection depends on the level of
exposure researchers have with the infected animals before the contamination is detected.
Exposure of Laboratory Workers.
Malicious acts involving deliberate or accidental exposure of laboratory workers could result in GoF virus
escape if the exposed individual(s) gets infected with the GoF virus. Human infection may occur with
virus from experimental samples, thawed virus, or fomites. If equipment or other materials are
deliberately contaminated or tampered with and laboratory workers are not protected well (i.e., through
use of the appropriate personal protective equipment), they may get infected with GoF virus in
experimental samples. Consequently, the likelihood of human infection is moderate.
Removal of GoF Virus from the Laboratory.
By definition, theft of GoF virus from the laboratory results in viral escape. However, the degree to which
GoF virus removal causes human infections depends on the form of the virus (i.e., either frozen virus or
virus in experimental samples) and/or the skills, expertise, and resources of the malicious actor to grow or
manipulate frozen viruses. Consequently, the likelihood of human infection is moderate.
Infection of Wild or Domestic Animals following Deliberate Outdoor Release of GoF Virus.
By definition, release of stolen GoF virus from experimental samples, stocks grown from stolen virus, or
stolen infected animals into the wild or households results in viral escape. The likelihood that a malicious
insider would be able to make a sophisticated dispersal device and not be detected is low, suggesting that
rudimentary dispersal devices may be the most likely route of release of virus. Furthermore, the
likelihood that infected animals (domestic or wild) could cause immediate infection in humans is low
because of the low level of interaction between wild animals and humans or domestic animals in urban
settings. However, the zoonotic nature of the viruses (i.e., their ability to infect animals and at least some
humans) does not automatically rule out the possibility of human infection ever.
Infection of Laboratory Workers or the Public following Deliberate Outdoor Release of GoF Virus
Exposure of laboratory workers or members of the public using stolen GoF virus from experimental
samples or stocks grown from stolen virus results in GoF virus escape and human infection.
7.4.1.3.1 Type of Breach Conclusion
The most likely types of breach leading to human infection following a malicious act are: release of
infected animals, infection of laboratory workers following deliberate release of GoF virus, and infection
of the public following deliberate release of GoF virus. However, successful release depends on form of
the virus (i.e., frozen stock or experimental sample) and the skill-level of malicious actors (to grow virus
from frozen stock). These breaches could only occur with the assistance of an insider or significant blast
that affects the integrity of the laboratory.

Given the right circumstances, human infections might occur from release of infected animals from and
within laboratories, cross-contamination of laboratory animals, exposure of laboratory workers, and
removal of GoF virus from the laboratory. The number of people exposed in each of these cases is likely
to be low, suggesting an even lower rate of infection among exposed individuals.390
7.4.1.4 Plausible Threats of GoF Viruses
The most plausible threats facing laboratories in which GoF virus research is stored or studies are those
carried out by insiders, acting alone or in cooperation with a domestic terrorist group or extremist group.
Insiders acting alone may be disgruntled, emotionally disturbed, or radicalized. Those cooperating with a
group may be sympathetic to the group’s cause, coerced, or subverted.
Most likely, insiders will commit acts covertly. Such acts would most likely expose a small number of
people to GoF virus. If exposed individuals are familiar with the symptoms and disease progression of the
viruses, they might seek help immediately if infected. If not (i.e., the general public), infections resulting
from exposure could lead to secondary infections. In addition, insiders could use GoF virus to expose a
large number of people.
Though less plausible, insiders might commit overt acts, such as arson, bombing, or armed assault. Some
of these acts would not lead to GoF virus escape and human exposure. The assumption is that emergency
responders and public health officials will respond quickly to overt acts involve active shooters, fire, or
explosions.
Most acts involving malicious actors without insider-assistance are not plausible. However, outsiders,
including transnational terrorists, domestic extremists, domestic terrorists, and lone outsiders, could carry
out an armed assault or detonate a bomb at the building perimeter if they have access. Armed assault
would not lead to GoF viral escape and human exposure. However, a bomb of sufficient size might affect
laboratory operating systems, possibly leading to release of GoF virus from experimental samples. These
acts are overt and would elicit response from emergency responders.
Table 7.6 summarizes the results of the analysis. These results provide the basis for epidemiological
modeling of plausible security threats involving GoF virus.
Table 7.6. Plausible Threats Involving High Containment Research Laboratories that Store or Study GoF
Viruses
Active shooter or physical assault
Insider
Overt Bomb detonated near or inside high containment space
Outsider Bomb detonated at building periphery
Covert Act
Removal of GoF virus (frozen stock or experimental sample), infected animals, or
(Expose Insider
contaminated equipment
Public)
Removal of GoF virus in experimental samples
Covert Act Deliberate contamination of personal protective equipment or laboratory
(Expose equipment
Insider
Laboratory
Workers) Deliberate compromise of laboratory equipment or personal protective equipment
Mixing of experimental samples or animals into lower containment
390 No known virus has 100% infection rate among exposed individuals.

In addition to these plausible threats, theft of information about research, facilities, hours of operation,
and personnel records are likely by foreign intelligence or domestic extremist groups.
7.4.2 Semi-Quantitative Epidemiological Modeling of Security Risks
7.4.2.1.1 The need for a semi-quantitative approach
The section above identified malicious acts that could plausibly be caused by a malicious actor and lead
to a loss of containment event. The variability in the manner through which these malicious acts could be
executed and the unknown probabilities of success at each step precludes the designing of fault trees (as
was done for accidents in the Biosafety Risk Assessment) for these malicious acts. That is, no evidence-
based quantitative model can be designed to estimate the probability that a particular malicious event
would be successful the amount of virus escaping containment from a successful malicious act. Moreover,
the state of the threat information is such that even estimating the frequency with which these malicious
events would be attempted would prevent open and transparent communication of the risks. For this
reason, a semi-quantitative approach is leveraged that estimates the difference in consequences between a
malicious act targeting a laboratory with wild type strains vs one targeting a laboratory with various GoF
strains, assuming that the malicious act were successful in causing at least one initial infection. This
section culminates with an estimate of the frequency with which these malicious acts must be successful
for the biosecurity risk to approximate the biosafety risk (given the relative consequences of the two types
of events). Throughout, the consequences computed from various events in Chapter 6 (Biosafety Risk) are
used where appropriate.
In most cases, any GoF trait would increase risk by either increasing the chance that an outbreak, initiated
by an infection caused by a malicious event, would escape local control to seed a global outbreak or by
increasing the consequences of a global outbreak. Two GoF traits theoretically could influence the chance
that an initial infection outside the laboratory would occur due to the malicious act: 1) enhanced growth in
culture (increasing the amount of contamination that could escape the laboratory); and 2) adaptation of
avian influenza strains to mammals so that the median infectious dose is decreased.
The first part of this section evaluates the potential for these two phenotypes to influence the probability
that an initial infection occurs. The sections that follow discuss how all other GoF phenotypes could
affect risk of an outbreak should an initial infection occur.
7.4.2.2 Influence of GoF traits on the probability that an infection outside the laboratory would occur
from a malicious act
The phenotypes of enhanced viral growth in culture and adaptation to mammals have the potential to
increase infection probability in loss-of-containment incidents. Of all of the pathogens assessed in this
study, this section is relevant only to influenza. Coronaviruses are already adapted to human hosts and so
this phenotype is meaningless for these pathogens. Moreover, coronaviruses already grow to high titers
and for this reason, no GoF manipulation is necessary to enhance their growth. (In any case, should a
scientist attempt to enhance their growth or decrease their infectious dose in people, the analysis herein
would suggest that little biosecurity would inhere in these manipulations).
Figure 7.6 explores the relationship of the amount of contamination released (which is influenced by the
titer of the sample leading to the contamination) in two strains of influenza, one with a relatively high
median infectious dose (like avian influenza—top panel) and one with a very low infectious dose (of
1.5pfu-bottom panel). Increasing the amount of pathogen escaping the laboratory by an order of
magnitude increases the probability of at least one infection by roughly 10%. That is, if a strain grew to a
100-fold greater titer due to a GoF manipulation and that concentrated stock caused contamination that

left the laboratory in a malicious act, the act would have only a 20% increase in the chance that an
infection would occur. Unless a very little amount of contamination leaves the laboratory (less than 100
pfu), this increase in the contamination released would increase the overall chance that an infection occurs
by less than a factor of two. Moreover, the analysis shown in Figure 7.6. assumes that an enhancement of
viral growth leads to a similar increase in the contamination released. In reality, viruses that grow to a
high titer are diluted for use in most experiments (such as plaque assays or challenge experiments) so only
some cultures that could cause contamination would be at the greater concentration enabled by a GoF
experiment.

Figure 7.6. The probability that a contamination event with wild type avian influenza (Panel A, above), or a
modified strain with a median infectious dose of 1.5pfu (Panel B, below) causes a certain number of infections
when contaminated material escapes the laboratory for a variety of viral loads. The y-axis shows probability
on a log scale with 1=100% certainty.
Comparing the panels in Figure 7.6. shows that the adaptation of avian influenza strains to mammals
(resulting in a lower median infectious dose) would increase the probability that at least one person was
infected by the contamination event by a factor of two or three. If just one pfu contaminates someone
leaving the laboratory, one person would be infected about 7% of the time if the strain had a low
infectious dose, compared to 2% if the strain were not adapted to humans. If the contamination involved

100 pfu, the chance of at least one infection would increase from 20% to 40% if the infectious dose were
to decrease. However, strains adapted to humans are likely to be minimally infectious to birds and
therefore malicious acts that involve the contamination of wild life (such as the release of infected birds)
would be a lower total risk if this manipulation occurred. If the avian influenza strain were not also
manipulated to be transmissible among humans, a malicious act resulting in the escape of contamination
would sicken at most five people (Figure 7.6.). In contrast, a strain that is not adapted to humans could
cause an outbreak in avians (if an infected animal were released, for example), which would lead to much
more severe human health consequences (up to 1,000 illnesses and 100 deaths) than the direct infection of
laboratory workers or the public by contamination. Recall that the biosafety risk assessment estimated
that adaptation of an avian strain to humans (without increasing transmissibility) would decrease risk by a
few fold because accidents that lead to an avian outbreak are much more likely than those that infect a
person and avian outbreaks could lead to the deaths of many people, not just a single person infected in a
laboratory. For this reason, because biosecurity events could lead to the infection of people or wildlife, we
presume that adaptation to humans neither significantly increases nor decreases biosecurity risk.
Events that involve the release of contamination from the laboratory could also result in the infection of
birds, although that chance is remote. Specifically, even assuming that 10E8 pfu of avian influenza
escapes the lab on a single person’s hand, the fomite model predicts that no infections would occur in
chickens, ducks or turkeys in 300,000 simulations. This result is not surprising because of the short half-
life of influenza on the skin (on the order of minutes) and the rarity of laboratory workers physically
handling poultry outside of a laboratory. No GoF phenotype would make the infection of birds from such
a contamination event more likely or more extensive should it occur because wild type strains of avian
influenza are already highly contagious and highly pathogenic in birds. The human health consequences
from such an outbreak are estimated to involve 100 deaths and 1,000 illnesses.
7.4.2.3 Aligning the malicious acts to biosafety scenarios to calculate risk of wild type agents
The probability that a malicious incident results in an outbreak that escapes local control depends heavily
on who is initially infected (a laboratory worker or a member of the public), if the infection is related to
an overt or covert incident and, to a lesser degree, how many people were initially infected. Events that
initially target laboratory workers are of lesser risk than those that target the public because laboratory
workers are more likely to be vaccinated against the strains in their laboratory, to self-monitor for initial
signs of illness (like a fever), and to be isolated should an unusual illness manifest. If the event is overt
and poses a high risk of causing an infection, the laboratory worker (or any person responding to an event
in a laboratory) could be given antivirals prophylactically and also would be more closely monitored for
the initial signs of illness (and could even be preemptively isolated).
Simply put, because of these four critical parameters, all malicious acts can be grouped into four
categories to consider the risk of a global outbreak of a human transmissible disease: overt infections of
laboratory workers, overt infections of members of the public, covert infections of laboratory workers,
and covert infections of members of the public (Figure 7.7). For diseases that can spread amongst wildlife
but not people (specifically, wild type avian influenza), a fifth group, malicious acts specifically infecting
wildlife, is considered.

Figure 7.7. Alignment of the various malicious acts with five categories of events considered in the biosafety
risk assessment. Because coronavirus strains that are modified to infect other animals pose a limited risk to
people, we do not consider theft of animals is likely to cause an outbreak of human-transmissible disease.
7.4.2.4 Probability that a malicious event would lead to a global outbreak
From the Biosafety Risk Assessment in Chapter 6, we can determine the probability that an outbreak,
caused by a malicious event, would lead to a global outbreak (Table 7.7). The table shows the probability
that at least one secondary case would be caused and the probability that the outbreak would escape local
control if laboratorians were infected or if members of the public were infected. Clearly, these
probabilities are greatly influenced by how many people were infected by the initial event. Table 7.7
shows how these probabilities change if just one person were initially infected or several people were
initially infected by the event. We assume that only five people in a laboratory could be simultaneously
infected by an event due to the relatively small numbers of people working in a containment suite at any
given time.

Table 7.7. Probability that the initial cases lead to secondary infections, and the probability that the outbreak escapes local control for each type of
event.
Event Type Seasonal Influenza Pandemic Influenza Avian Influenza Coronaviruses
Risk Type # of Prob. of Prob. Prob. of Prob. Prob. of Prob. Prob. of Prob.
Initial secondary escapes secondary escapes secondary escapes secondary escapes
Cases spread local spread local spread local spread local
control control control control
Infection of wild
N/A N/A N/A N/A 100% 100% N/A N/A
life
Covert infection 1 55% 20% 60% 30% 0% 0% 35% 0%

of public 10 100% 90% 100% 95% 0% 0% 100% 0%
Covert infection 1 15% 20% 20% 30% 0% 0% 2% 0%

of worker(s) 5 85% 90% 85% 95% 0% 0% 32% 0%
Overt infection 1 30% 20% 40% 25% 0% 0% 25% 0%

of public 10 100% 90% 100% 90% 0% 0% 100% 0%
Overt infection 1 1% 20% 1% 25% 0% 0% 1% 0%

of workers 5 3% 90% 3% 90% 0% 0% 5% 0%
Events involving wild-type viruses highlighted in yellow have a relatively low probability of causing a pandemic and therefore offer an opportunity for GoF
manipulations to increase risk.

As can be seen from the table, events that lead to at least one infection in members of the public are five-
to 30-fold more likely to initiate a local outbreak than those that infect laboratory workers. Events that
covertly infect laboratory workers are 10- to 20-fold more likely to initiate a local outbreak than those that
overtly infect laboratory workers. This analysis permits a relative ranking of the risk (assuming the
probabilities of the relative chance of success and the frequency of the malicious acts are unknowable),
which is shown in Table 7.8.
Table 7.8. Relative risk of the plausible malicious acts given an unknown frequency of occurrence and
probability of success.
Risk Category Primary Infection Overt vs Covert Event
Highest Public or Wildlife Covert Theft of animals
Highest Public Covert Theft of equipment, theft of virus
Moderately High Public or Wildlife Overt Bombing
All events that lead to the infection of co-
Moderately Low Laboratorians Covert
workers directly or indirectly
Lowest Laboratorians Overt Shooting
Events that lead to an infection of wildlife are relatively low risk because those cannot seed a global pandemic of a
human transmissible disease.
The potential for GoF phenotypes to increase the probability that a global outbreak occurs following an
infection initiated by a biosecurity event is taken directly from the Biosafety Risk Assessment (Chapter 6
and summarized in the Stop Light Chart shown in Figure 7.8). This figure shows that transmissibility is
the trait that can most affect the probability that the outbreak would escape local control, and this
statement holds true for all pathogens evaluated except for pandemic influenza (which is sufficiently
transmissible so that further increases influence the probability of an outbreak escaping only minimally).
In fact, for pandemic influenza, no GoF trait significantly increases the risk that an outbreak would spread
globally. If the strain were more pathogenic, perhaps the public fear elicited would improve social
distancing measures and decrease the probability that an outbreak is contained, but this possibility cannot
be directly evaluated. The ability to overcome protective vaccination and antiviral resistance
independently modestly increases the chance that an infected laboratory worker would cause a secondary
infection, so this trait has minimal influence on risk. Lastly, no explicit plans exist for the extensive use of
antivirals in an outbreak associated with a laboratory, so the role of antivirals in a nascent outbreak could
not be determined.

GoF Seasonal Influenza Pandemic Influenza Avian Influenza Coronaviurses
Phenotype Viruses Viruses Viruses
Enhanced
transmissibility
Enhanced
Unknown Unknown
pathogenicity
Adaptation to
N/A N/A N/A
mammals
Evasion of
induced N/A N/A
immunity
Evasion of
natural/residual N/A N/A
immunity
Antiviral
N/A
resistance
Enhanced
growth in N/A N/A N/A N/A
culture/eggs
Figure 7.8. A chart showing the relative increase in the probability that a global outbreak would occur for a
variety of pathogens with GoF traits compared to the same strains with wild type traits. Darker orange
denotes increasing risk. Green indicates that the phenotype does not increase risk for that pathogen.
In contrast, for seasonal influenza, the ability to evade residual immunity or an increase in transmissibility
to that of pandemic influenza would increase the probability of a global outbreak by two to three fold.
The relatively low risk that an infected laboratorian would infect another person is due to robust health
monitoring and isolation protocols. GoF traits do not reduce the ability of these measures to mitigate an
incident.
Since wild type avian strains are not transmissible among people, the hazard ends with those initially
infected by the event unless wild birds are infected (causing a global avian outbreak) or the strain is
modified to transmit among humans. If the strain were modified to be as transmissible in humans as
seasonal or pandemic influenza, the risk of a global outbreak would be significant. For this reason, GoF
studies that increase the transmissibility of avian strains in humans significantly increase the probability
that a global outbreak would occur. No other GoF traits affect the probability that an outbreak seeds a
global pandemic for avian influenza.
Similarly, the coronaviruses are insufficiently transmissible to have a significant chance of seeding a
global pandemic. Strains with enhanced transmissibility increase the chance that an outbreak occurs and
that this outbreak sparks a global pandemic.

7.4.2.5 The influence of GoF on the consequences of a global pandemic
Should a global outbreak be sparked by a malicious act targeting a laboratory, the consequences would be
similar to a global outbreak sparked by an accident in a laboratory and the influence of GoF traits on risk
would be identical to those explored in Chapter 6. Figure 7.9 summarizes those findings.
GoF Phenotype Seasonal Pandemic Avian Influenza Coronaviruses

Influenza Viruses Influenza Viruses Viruses
Enhanced
transmissibility
Enhanced
pathogenicity
Adaptation to
N/A N/A N/A N/A
mammals
Evasion of
N/A N/A
induced immunity
Evasion of
natural/residual N/A N/A
immunity
Antiviral
N/A
resistance
Enhanced growth
N/A N/A N/A N/A
in culture/eggs
Figure 7.9. A chart showing the relative increase in the probability that a global outbreak would occur for a
variety of pathogens with GoF traits compared to the same strains with wild type traits. Darker orange
denotes increasing risk. Green indicates that the phenotype does not increase risk for that pathogen.
For seasonal and pandemic influenza, antiviral resistance and the ability to overcome protective
vaccination would not significantly increase deaths from an outbreak globally, but would increase deaths
by a few fold in North America due to the availability of these countermeasures in the U.S. Note,
however, that to effectively evade the protection afforded by a vaccine raised in response to a particular
outbreak, the strain must be modified to overcome vaccination regardless of its antigenic profile, which is
not a subject of active GoF research. No other trait influences the global consequences of a strain of
pandemic influenza significantly.
For seasonal influenza, increasing the transmissibility (or, similarly, imbuing the ability to evade residual
immunity) increases global deaths by a few fold. Given the relatively low case fatality rate of seasonal
influenza, significant increases in pathogenicity (10x or more) are possible and these would
proportionally increase the death toll.
A wild type avian influenza strain can infect people only via contact with infected birds, resulting in a few
thousand cases at best. Given that many strains are minimally pathogenic, increasing the pathogenicity in
people could increase these deaths by a few fold. In contrast a strain modified to be transmissible in

people could cause a global outbreak, infecting millions and therefore significantly increasing risk.
Increasing pathogenicity could increase global deaths by a few fold.
The wild type versions of the coronaviruses are insufficiently transmissible to have a significant
probability of causing a global outbreak, or, if they do, the consequences are relatively small. Increasing
transmissibility of these strains
7.4.2.6 Overall influence of GoF on risk of biosecurity events
In summary, only a handful of GoF traits significantly increase biosecurity risk after a malicious event
targets a laboratory. For seasonal and pandemic influenza, the ability to overcome protective vaccination
and antiviral resistance modestly increases risk by increasing the potential consequences in North
America. No significant effect on risk exists if the global population is considered as a whole. No other
trait significantly influences the risk of pandemic influenza strains.
For seasonal influenza, increasing the transmissibility and ability to evade residual immunity significantly
increases risk because outbreaks are more likely to occur, escape local control and create more
consequential global outbreaks.
For avian influenza, increasing transmissibility greatly increases risk because this modification is required
to spark a global outbreak of a disease by human-to-human contact, potentially infecting millions.
Without this change, the hazard is restricted to those exposed to contaminated materials and infected
birds, limiting the outbreak to thousands of cases at most. Increasing pathogenicity can modestly increase
risk.
7.4.2.7 Comparison of risk of biosecurity events vs biosafety events
To understand the biosecurity risk of acts targeting a GoF laboratory relative to the risk of accidents with
the same pathogens, this section provides data on the approximate frequency that various malicious acts
must successfully result in an infection to match the risk of an accident involving the same pathogen. To
accomplish this, estimates of the probability that a laboratory acquired infection sparks a global pandemic
from the Biosafety Risk Assessment in Chapter 6 are combined with historical rates of laboratory
acquired infections. Figure 7.10. shows the return frequency of a laboratory acquired infection in any one
of the approximately 100 laboratories that study influenza or the coronaviruses in the US given that no
laboratory infections have occurred in the last 20 years (or assuming that a few have occurred that we
have not identified).

Figure 7.10. The predicted return period of laboratory acquired infections (LAIs) assuming 0-10 infections
have actually occurred in the last 20 years across 100 laboratories. The 90 th percentile of the maximum rate
(bottom line) was used to produce an estimate of the return period that would greater than 90 out of 100
estimates of the frequency, whereas the maximum likelihood estimate and 90 th percentile of the minimum rate
(top line) is also shown.
A Laboratory acquired infection is expected to occur every three to 200 years across all laboratories in the
U.S. For simplicity, all these infections are assumed to be in laboratories that study seasonal influenza,
since these vastly outnumber the laboratories that study other pathogens and this work can be done at
BSL2, which allows more laboratory acquired infections to occur compared to BSL3. (Highly pathogenic
avian influenza, SARS-CoV, and MERS-CoV are studied in BSL 3 laboratories suggesting that the
calculated number represents the upper bound of laboratory acquired infections for these agents.)
As described in the Biosafety Risk Assessment (Chapter 6), only about 0.5% of these laboratory
infections are predicted to cause global pandemics due to public health response measures, stochastic
factors, health monitoring, and isolation protocols. For this reason, a global pandemic due to a laboratory
accident is expected to occur every 750-50,000 years.

Given that the highest risk biosecurity events (theft of animals, materials or stocks by an insider) are also
among the most plausible, and that these events lead to covert infections of the public, the chance that a
biosecurity event that infects one person leads to a global pandemic is much greater than the chance of an
accidental laboratory acquired infection (since these may be overt and nearly always infect laboratory
workers). If an infection occurs, biosecurity events have a 11% chance of starting a global pandemic
(55% chance of initiating an outbreak and a 20% chance that this outbreak escapes local control).
For a biosecurity event to have the same total risk as biosafety events, a successful event that covertly
infects the public (theft from an influenza laboratory of an infected animal, contaminated piece of
equipment or viral stock) must occur once every 80-5,500 years (11% of 750-50,000). Given the
frequency with which thefts have been perpetrated by insiders in laboratories, this analysis suggests that

8 Biosecurity Risk of GoF Information
Summary 212
Purpose and Approach 214
Methods 214
8.3.1 Use of Sources 214
8.3.2 Methodology for baselining the biological threat 214
8.3.3 Methodology for baselining the state of the science 215
8.3.4 Evaluation of the Capability and Intent of Malicious Actors to Leverage Dual Use
Information 215
Baselining the Biological Threat 216

8.4.1 Mortality from Diseases Caused by Non-GoF Pathogens 217
8.4.2 Incapacitation from Diseases Caused by Non-GoF Pathogens 219
8.4.3 Footprint of Attacks of Non-GoF Pathogens 220
8.4.4 Gain of Function Strains Compared to Naturally Occurring Strains 224
Overview of the State of the Science of Dual Use GoF Information 228
8.5.1 State of the Science of GoF Experiments in Influenza Viruses 228
8.5.2 State of the Science of GoF Experiments in the Coronaviruses 235
8.5.3 Overview of the State of the Science of GoF Experiments 236
Evaluation of the Capability and Intent of Malicious Actors to Leverage Dual Use Information 237

8.1 Summary
In this section, we analyze the risk that a malicious actor might misuse the information in publications
describing GoF research. This analysis is based on the open-source literature covering desirable
characteristics of biological agents and the scientific literature on GoF studies and non-GoF studies with
significant dual-utility. We employed the NSABB definition of GoF research to delineate the dual-use
phenotypes considered.391
We assessed the potential biosecurity information risk that could be generated by GoF information
compared to what could be achieved through dual-use studies that do not rely on GoF research. We then
assessed whether the unique dual-use information resulting from GoF studies had already been published.
We find that little information risk remains from GoF research (see Figure 8.1). Although the
development of a highly-contagious, highly virulent strain of influenza presents significant biosecurity
information risk, the methods to produce these strains have already been published and so no information
risk remains. Moreover, the specific changes in the genome that led to these traits have also been
characterized and published, so an actor could reproduce the dual-use strains using reverse genetics.
Although several potentially dual-use studies have already been published, translating animal studies of
transmissibility to empirically predict an exact R0 in a human outbreak is currently impossible; therefore,
we cannot determine if the studies already published could be used to create strains of influenza that
could cause a global pandemic (R0 of greater than one). If not, further studies on this topic could create an
information risk.
Similarly, information on how to develop strains of influenza viruses that grow well in culture/eggs or
evade medical countermeasures or diagnostics has some dual-utility, but the methods to create these
strains also have already been published.
rch-APPROVED.pdf

Dual-Use GoF Phenotype Seasonal/Pandemic Influenza Coronaviruses
Enhanced transmissibility in
mammals
Published methods require skills in
molecular biology or were in poor
Enhanced pathogenicity in animal models of pathogenicity.
mammals No publications exist on creation
of influenza strains that lead to
chronic illness.
Enhanced transmissibility while
maintaining pathogenicity
Overcoming natural or induced Via the creation of antigenically
N/A
immunity distinct strains only
The evasion of diagnostics that target
Evading diagnostics the genomic sequence of the virus
may pose an information risk.
Enhanced production in cell
N/A
culture or eggs
Figure 8.1. Summary of the information risk posed by GoF research in influenza (middle set of columns) and
the coronaviruses (right set of columns). Information that has a significant dual-use (from Figure 8.2) AND is
not yet published (Figure 8.3) is shaded darkly because it poses a remaining information risk. Information
that is not actually dual-use OR has already been published is left white because it poses no remaining
information risk. Information shaded gray may have some remaining information risk under some
circumstances. N/A denotes traits that are not applicable to the coronaviruses.
Significant information risk would be realized by the publication of methods to create a highly
transmissible SARS- or MERS-coronavirus that maintains its pathogenicity. Notably, without an animal
model of transmissibility for these pathogens, this information risk is unlikely to be realized in the near
future. A modest information risk inheres in methods to manipulate the genomic targets of a diagnostic
assay for coronavirus infections without compromising the other desirable traits of the pathogen.
A modest information risk would be realized if researchers published methods to produce strains of
influenza viruses that can produce more prolonged or chronic illness. Although this manipulation is a
possible enhancement of pathogenicity that can fall under the definition of GoF research, there is little
scientific rationale to undertake these experiments. Hence, the possibility that this information risk will be
realized is low. Another modest information risk inheres in the publication of methods to produce strains
of influenza virus that are able to overcome protective vaccination even if the vaccine matches the
serotype of the pathogen. Similar work has been published for other pathogens, but these pathogens have
larger and more plastic genomes than the influenza viruses so it is not known if similar manipulations
could be successfully carried out in the influenza viruses.
State actors (and the sub-state groups they sponsor) are currently the only groups with the resources,
expertise, motivation and time to leverage this dual-use information. These states could protect their own
populace from a global pandemic by secretly stockpiling vaccines that are protective against their
modified strain. For this reason, states would be more likely to produce modified influenza viruses than
coronaviruses (because no vaccines exist for this type of agent) and would probably be uninterested in
developing strains able to overcome any vaccine (as this strain would vitiate their comparative
advantage). Sub-national malicious actors may obtain the capability to replicate some of the less complex

GoF studies, but have so far not demonstrated any capacity to work with viral agents and little capacity
for waging biological warfare in general. Highly skilled individuals trained in biology would be capable
of replicating GoF studies, but are currently constrained greatly by a lack of material resources and time
Finally, no information risks unique to GoF research were identified. Similar techniques to those used in
GoF experiments could be leveraged for other pathogens to create a highly transmissible strain of an
already deadly virus (like the Hendra and Nipah viruses) or to create a deadly strain of an already highly
transmissible pathogen that has been modified to overcome protective vaccination (polio-, mumps or
measles-virus). Perhaps most worryingly, reverse genetics techniques could be used to synthesize
smallpox virus if an actor has significant molecular biology skill, and this strain could be modified to
overcome protective vaccination. Non-GoF pathogens could be used to produce effective, novel
incapacitating agents by the modification of a highly contagious virus (polio-, mumps- or measles-virus)
to overcome protective vaccination.
8.2 Purpose and Approach
The purpose of this task is to identify those GoF studies on influenza, SARS, and MERS viruses that, if
published, would provide useful information to a malicious actor seeking to create a biological weapon.
This analysis assumes that the body of dual-use information already in the public domain is significant,
and so seeks to identify studies that would contribute to the ability of a malicious actor beyond what has
already been published. Since an adversary is presumably interested in causing harm in any way possible,
this analysis considers GoF studies on influenza, SARS-CoV, and MERS-CoV in light of what can
already be achieved with unmodified strains of these pathogens and non-GoF pathogens. Indeed, the
capability to cause harm with agents other than influenza virus, SARS-CoV and MERS-CoV is
significant. Hence, this comparative assessment must be conducted to understand the advantage an
adversary gains by leveraging the information gleaned from GoF studies, specifically. Lastly, to provide
insight into the possibility that novel, dual-use information would be exploited if it were published, this
study examines the capability and motivation of malicious actors to weaponize pathogens.
8.3 Methods
8.3.1 Use of Sources
This biosecurity information risk assessment involves the analysis of the biosecurity risk posed by the
future publication of GoF research results beyond the existing dual-use information already in the public
domain. This analysis uses scientific data to identify potential new capabilities afforded by GoF research
to those who seek to cause harm. Biomedical literature describes the infectiousness, pathogenicity and
countermeasure resistance of wild type pathogens, and potential modifications to pathogens to enhance
any of these traits. Information from intelligence/law enforcement data was used to provide the general
context necessary to understand the capabilities of malicious actors to exploit this research but could not
be directly reported at an unclassified level. Beyond this contextual level of discussion, we relied on
open-source information on offensive biological weapons programs undertaken by states and non-state
actors to source our analysis of malicious actor intent and capability.
8.3.2 Methodology for baselining the biological threat
We first conducted an analysis of the biomedical literature and open-source descriptions of state-
sponsored offensive weapons programs to determine what a malicious actor using unmodified agents
could achieve. We then examined how GoF pathogens could provide additional capabilities to an

adversary. In this analysis, we considered the ability of various pathogens to incapacitate and kill as the
possible desired outcomes of a biological attack. Attacks targeting animals for the purpose of causing
economic harm or harm to animal health were outside the scope of the assessment. We also considered
the “footprint” of the attack, meaning the area and time over which the attack would incapacitate or kill
(under the assumption that a larger area or time of effect was desirable). Contagiousness of GoF
pathogens is considered in this context. Given this baseline, our analysis identifies the type of information
created via GoF studies that would prove useful to adversaries seeking to build additional biological
weapon capabilities.
One quantitative method was used to baseline the threat. To quantitatively assess the dual-utility of the
phenotype of enhanced growth, we compared how the number of victims infected from an intentional
release scaled with the total amount of pathogen aerosolized, which itself is a function of how much
pathogen can be produced. To perform this assessment, we used the Hazard Prediction and Analysis
Capability (4.0) as described in the Risk Assessment of Accidents and Natural Disasters section above.
We modeled only New York City as the target (due to its population density) across 12 different weather
conditions for each release amount to show the maximum extent of a large attack.
8.3.3 Methodology for baselining the state of the science
Given that very little can be done about the dual utility of studies already published, we characterized the
state of the science regarding the enhancement of all traits described in the NSABB framework. We
analyzed the body of literature that encompasses all GoF studies identified by the project team for the
benefit assessment and/or risk assessment (see bibliography). Specifically, we sought to understand to
what degree the methods for the creation of modified strains of influenza viruses and coronaviruses with
the following phenotypes already exist in the public domain:
• Enhanced production of pathogens in vitro or in ovo (high titer),

• Enhanced mortality,
• Enhanced morbidity,
• Evasion of natural or induced immunity, and
• Evasion of medical countermeasures, including vaccines, antivirals and diagnostics.
This task culminated with the identification of GoF research that would provide uniquely valuable
information to a malicious actor for misuse beyond the body of dual-use research that already exists.
Also, we identified whether dual-use information already in the literature requires a particularly
challenging technical approach in order to ascertain if a biosecurity information risk could be suffered via
the publication of an easier experimental route to the same product. Similarly, instances in which the
researchers published the specific genetic changes leading to the desired traits are noted because a
malicious actor could simply recreate the useful strains using reverse genetics instead of repeating the
methods. This section highlights which of the phenotypes described under the funding pause have yet to
be achieved in the published literature, representing a remaining, possible information risk.
8.3.4 Evaluation of the Capability and Intent of Malicious Actors to Leverage Dual Use Information
We used open-source information to characterize the technical skill, sophistication and resources required
to replicate those GoF experiments that provide information uniquely useful and of interest to a malicious
actor. We relied on historical precedent, as documented in open source information, in considering

whether certain malicious actors might have intent to leverage uniquely dual use information yet to be
published.
8.4 Baselining the Biological Threat
When considering the information produced by GoF experiments, we considered how the results achieved
intersect with the goals of those wishing to misuse the information. As in other sections of this report, we
used the NSABB definition of GoF research for this analysis.392 Specifically, we consider various strains
of seasonal, pandemic and avian influenza and the MERS and SARS coronaviruses. The phenotypes we
consider are:
• Enhanced production of pathogens in vitro or in ovo (high titer),

• Enhanced mortality,
• Enhanced morbidity,
• Evasion of natural or induced immunity, and
• Evasion of medical countermeasures, including vaccines, antivirals and diagnostics.
From the perspective of an adversary seeking to create a biological weapon (called a “weaponeer”) these
phenotypes can be described by three agent/weapon characteristics. Mortality covers the GoF phenotype
of enhanced mortality and the ability of a pathogen to evade medical countermeasures and natural or
induced immunity, as its ability to do so increases the overall case fatality rate. Incapacitation covers the
GoF phenotype of enhanced morbidity, but also the phenotypes describing the evasion of medical
countermeasures and natural or induced immunity, as these abilities increase the attack rate or the severity
or duration of illness. Footprint—the ability of a weapon to cover an area, extend a pathogen’s effects
over time or its ability to reach a set number of victims—encompasses several GoF phenotypes. A strain
with enhanced production characteristics can be used to increase the effective payload of a weapon (i.e.
the same production run can produce more pathogen), potentially infecting more victims and covering a
larger area when the agent is released using a weapon. A highly contagious GoF strain increases the
footprint of an attack by increasing the number of victims harmed after the primary aerosol, which, in
turn, increases the geographic and temporal extent of the effects. Similarly, a strain that evades medical
countermeasures increases the number of victims potentially harmed by the primary aerosol. For
contagious strains, the evasion of medical countermeasures also increases the attack rate and geographic
and temporal extent of a resulting outbreak compared to an outbreak that can be effectively controlled by
medical countermeasures.
To understand how GoF research could provide information that increases the ability of a weaponeer to
produce a weapon that is highly lethal, highly incapacitating or has a large footprint, we compare these
GoF outcomes with what is possible without GoF research. We first consider the phenotypes separately
and then consider under which circumstances the combination of traits leads to a particularly useful strain.
rch-APPROVED.pdf.

8.4.1 Mortality from Diseases Caused by Non-GoF Pathogens
Without any information from GoF research, a weaponeer can choose from several agents that cause
diseases with extremely high mortality rates, which can be identified simply by scrutinizing the Select
Agent list.393
8.4.1.1 The bacterial agents
Several diseases caused by bacterial agents have an extremely high case fatality rate. Inhalational anthrax
has an untreated case fatality rate of 90%, and a treated case fatality rate of approximately 50% if
aggressive treatment is provided.394 Melioidosis has a case fatality rate in western countries of about 15%,
although many of the victims have significant co-morbidities.395,396 Pneumonic plague is almost uniformly
fatal if untreated.397 Although the untreated case fatality rate of the typhoidal form of tularemia is about
30%, animal studies suggest that high doses that may be experienced in the context of a biological attack
significantly increase the lethality of this agent.398,399
Because all of these agents are bacteria that can replicate outside of a host cell, a weaponeer would likely
find isolating, growing and weaponizing these agents easier than the influenza viruses and
coronaviruses.400 Also, since many of the bacterial Select Agents featured in the offensive weapons
programs of several states, information on their efficient weaponization already could be available or
obtained by state actors.401,402,403
From a weaponeer’s perspective, the disadvantage of using bacterial agents is that the diseases they cause
can be prevented or effectively treated with antibiotics. However, simple molecular or microbiological
methods (such as selection in vitro or in vivo) can be used to induce significant resistance in these bacteria
to a panel of therapeutically useful antibiotics. Moreover, the methods to eliminate the fitness defect
associated with newly acquired antibiotic resistance (or indeed any newly acquired phenotype) also
involve relatively simple microbiological manipulations. In short, when compared to methods related to
393 US Government Publishing Office, “Title 42: Public Health, §73.3 HHS Select agents and toxins,”
http://www.ecfr.gov/cgi-bin/text-
idx?SID=a2b0afcad59ea49b88e4bf9e9b20a26c&mc=true&node=pt42.1.73&rgn=div5#se42.1.73_13.
394 Jon-Erik C. Holty et al., “Systematic review: a century of inhalational anthrax cases from 1900 to 2005,” Annals of Internal
Medicine 144 no. 4 (February 2006): p. 270-280, http://annals.org/article.aspx?articleid=720551.
395 Saïdani N, et al., “Melioidosis as a travel-associated infection: Case report and review of the literature,” Travel Medicine
and Infectious Disease (4 September 2015) http://www.sciencedirect.com/science/article/pii/S1477893915001428.
396 Nasner-Posso K, et al., “Human melioidosis reported by ProMED,” International Journal of Infectious Diseases 35 (June
2015): p.103-104, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4508390/
397 Kiersten J. Kugeler et al., “Epidemiology of Human Plague in the United States, 1900-2012,” Emerging Infectious Diseases
21, no. 1 (January 2015): p. 16-22, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4285253/.
398 Joseph R. Egan, Ian M. Hall, Steve Leach, “Modeling Inhalational Tularemia: Deliberate Release and Public Health
Response,” Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science 9, no. 4 (2011): p.334-335,
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3223019/pdf/bsp.2011.0004.pdf.
399 Glynn A, et al., (2015) “Comparison of experimental respiratory tularemia in three nonhuman primate species,”
Comparative Immunology, Microbiology and Infectious Diseases 39 p. 13-24,
http://www.ncbi.nlm.nih.gov/pubmed/25766142.
400 This argument was made in general form in: Jonathan B. Tucker, “Bioterrorism: Threats and Responses,” Biological
Weapons: Limiting the Threat, ed. Joshua Lederberg (Cambridge: The MIT Press, 2001), p. 286.
401 Leitenberg M, Zilinskas R, (2012) The Soviet Biological Weapons Program: A History Cambridge: Harvard University
Press
402 United Nations Monitoring, Verification and Inspection Commission (UNMOVIC), “Compendium: Chapter V, the
Biological Weapons Programme,” retrieved at:
https://web.archive.org/web/20131203182832/http://www.un.org/depts/unmovic/new/documents/compendium/Chapter_V.p
df.
403 The Secretary of Defense, “Memorandum For the President, National Security Decision Memoranda 35 and 44,” July 6,
1970, Declassified http://nsarchive.gwu.edu/NSAEBB/NSAEBB58/RNCBW22.pdf.

GoF, a weaponeer can much more easily obtain a highly lethal strain of bacteria via simple methods of
manipulation.
8.4.1.2 The viral agents
Several viral agents on the Select Agent list are associated with a high case fatality rate. The Hendra and
Nipah viruses have a case fatality rate of roughly 50%, with no effective treatment. 404,405,406,407 Marburg
virus, a hemorrhagic fever virus (HFV) on the Select Agent list that was weaponized by the Soviet Union,
had a 22% case fatality rate in Europe during its initial outbreak, and rates above 80% in subsequent
outbreaks in the developing world. 408,409 Hemorrhagic fever case fatality rates are worsened by the
difficulty of applying proper clinical care management and the lack of non-experimental treatments with
demonstrated efficacy.410Since Marburg HFV featured in the offensive biological weapons program of the
Soviet Union, a malicious state-level actor may be able to access the methods to magnify and weaponize
this agent.411 Although not a select agent, rabies virus causes a nearly uniformly lethal infection if not
prevented by vaccination soon after exposure. Finally, while currently circulating strains of H5N1 avian
influenza are associated with a fatality rate of 60%, this rate may be inflated due to potentially unreported
cases of mild illness in this outbreak. 412,413 SARS and
MERS outbreaks are associated with a case fatality rate Several bacterial and viral agents give the
greater than 10% as well, albeit mostly in the elderly (see weaponeer a choice of pathogens that are
Sec. 4).414 In short, a weaponeer can use a variety of wild highly lethal without relying on
type viruses if high mortality is desired without resorting to information from GoF experiments.
the exploitation of more sophisticated GoF methods.
8.4.1.3 Toxins
Several toxins are listed on the Select Agent list.415 These toxins are highly deadly and lack effective
treatments for victims who have received a sufficiently large dose. Extracting or otherwise producing
enough toxin from biological organisms to inflict a mass casualty requires industrial-like production
capacity. That being said, several state actors and one sub-state actor has invested in the capacity to
404 Broder C, et al., “A treatment for and vaccine against the deadly Hendra and Nipah viruses,”
405 Centers for Disease Control (CDC), “Hendra Virus Disease (HeV): Treatment,”
http://www.cdc.gov/vhf/hendra/treatment/index.html.
406 Playford E, et al., “Human Hendra Virus Encephalitis Associated with Equine Outbreak, Australia, 2008,” Emerging
Infectious Diseases 16, no. 2 (February 2010), http://wwwnc.cdc.gov/eid/article/16/2/09-0552_article.
407 Robin McConchie, “Hendra trials for humans about treatment not prevention,” ABC Rural, April 1, 2015,
http://www.abc.net.au/news/2015-04-01/human-hendra-drug-treatment-not-
prevention/6365472?WT.ac=localnews_brisbane.
Press.
409 Mehedi M, et al., (2011) “Clinical aspects of Marburg hemorrhagic fever,” Future Virology p. 1091-1106,
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3201746/.
410 Ippolito G, et al. (2012) “Viral hemorrhagic fevers: advancing the level of treatment,” BMC Medicine 10, no. 31
www.biomedcentral.com/1741-7015/10/31.
Press.
412 Bui C, et al. (2015) “A Systematic Review of the Comparative Epidemiology of Avian and Human Influenza A H5N1 and
H7N9- Lessons and Unanswered Questions,” Transboundary and Emerging Diseases p.6,
http://onlinelibrary.wiley.com/doi/10.1111/tbed.12327/references.
413 Morens D, Taubenberger J (2015) “How Low Is the Risk of Influenza A (H5N1) Infection?,” The Journal of Infectious
Diseases 211, no. 9 p. 1364-1366, http://jid.oxfordjournals.org/content/211/9/1364.long.
414 Guan Y et al Molecular epidemiology of the novel coronavirus that causes severe acute respiratory syndrome. The Lancet
363: 99-104
415 US Government Publishing Office, “Title 42: Public Health, §73.3 HHS Select agents and toxins.”

produce toxins in quantities useable in weapons.416,417 Therefore, for adversaries willing to invest in
industrial scale production of toxins, GoF information provides little value for achieving a highly lethal
agent because toxins are already very deadly.
8.4.2 Incapacitation from Diseases Caused by Non-GoF Pathogens
Some biological weapons are not designed to kill, but rather to incapacitate the soldiers or industrial
workers of an enemy. In fact, incapacitating agents featured heavily in the now defunct US offensive
biological weapons program.418 Due to their high case fatality rate, MERS and SARS coronaviruses
cannot be considered incapacitating agents and are not discussed further in this section. The effectiveness
of an incapacitating agent can be described by three characteristics: the infectious dose, the severity of the
symptoms, and the duration of incapacitation.
Wild type strains of the incapacitating agents weaponized in the former US offensive biological weapons
program, such as Coxiella burnetii, have a number of characteristics that make them suited for this
purpose. The median infectious dose in humans of C. burnetii is less than ten microbes, which is
comparable to the most infectious strains of influenza and the coronaviruses, and provides little
opportunity for improvement utilizing information from GoF studies.419 The symptoms of the acute
disease caused by infection with C. brunetii¸ called Q fever, are similar in severity and type to influenza,
including a high fever (up to 105°F), pain, headache, malaise, vomiting and diarrhea.420 These symptoms
persist longer than the symptoms of influenza, with fever typically lasting longer than 10 days (fever in
influenza lasts typically half as long—See Supplemental Information on the disease course of
influenza).421 Moreover, some victims develop a chronic form of Q fever with long-lasting and recurrent
disabling symptoms.422 Antibiotics can be used to effectively treat the illness, but, as described above,
antibiotic resistance can be imbued into this agent using methods much less technically challenging than
those necessary to undertake GoF studies. Moreover, because C. burnetii was weaponized in the former
US weapons program, the information needed to grow and weaponize this agent could be leveraged by an
adversary. 423,424 In short, GoF studies with the influenza viruses are unlikely to lead to the development of
a pathogen that is more effective as an incapacitating agent than C. burnetii because this agent is highly
infectious and produces a severe, relatively
long-lasting illness. The only caveat is that GoF studies with the influenza viruses are unlikely to
influenza infections have a relatively fast lead to the development of a pathogen that is better as an
symptom onset time compared to C. burnetii incapacitating agent than C. burnetii, unless rapid
infections (an average of 2 days for symptom onset times are desired by a malicious actor
above all other characteristics.
influenza infections, versus 2-3 weeks for
Biological Weapons Programme,”
Michael A. Guhin to Robert M. Behr, “Memorandum for Dr. Kissinger, Subject: The Toxins Issue.”
Press
418 See Tab A: Material to be Destroyed (Biological and Toxin), in:
The Secretary of Defense, “Memorandum For the President, National Security Decision Memoranda 35 and 44,” July 6,
1970, Declassified, p.3, http://nsarchive.gwu.edu/NSAEBB/NSAEBB58/RNCBW22.pdf.
419 Russell John Brooke et al., “Human dose response relation for airborne exposure to Coxiella burnetii,” BMC Infectious
Diseases 13, no. (2013): http://www.biomedcentral.com/1471-2334/13/488.
420 Centers for Disease Control and Prevention, “Q Fever: Symptoms, Diagnosis, and Treatment,” November 13, 2013,
http://www.cdc.gov/qfever/symptoms/.
421 Ibid.
422 Ibid.
423 The Secretary of Defense, “Memorandum For the President, National Security Decision Memoranda 35 and 44,” p.3.
424 William J. Broad, “US Selling Papers Showing How to Make Germ Weapons,” The New York Times, January 13, 2002,
http://www.nytimes.com/2002/01/13/national/13GERM.html.

acute symptom onset for C. burnetii infections), and a malicious actor that strongly values rapid effects
over other weapons characteristics may favor influenza. In this case, wild type influenza viruses may be
considered nearly ideal. 425
8.4.3 Footprint of Attacks of Non-GoF Pathogens
As described above, the footprint of an attack is defined as the number of victims affected, the physical
area affected or the duration of the disruption directly caused by the attack. In Sec. 8.4.3.1 below, we
explore how GoF phenotypes could affect the footprint of a weapon. In this section, we explore how wild
type pathogens and existing technology can create large footprint attacks.
The biological agent that most notoriously embodies attributes desirable for large footprint attacks is B.
anthracis, which led to its inclusion in several state offensive weapons programs.426,427,428 The spores
formed by B. anthracis are extremely resistant to environmental forces and can survive for a long time
suspended in an aerosol.429 This pathogen is able to create large footprint attacks, which is demonstrated
by the use of the related organism, B. thuringiensis, in pest control programs involving the treatment of
square miles of territory with spores dispensed from a single vehicle.430 If dispersed by sophisticated
maritime, ground-based or aerial platforms, B. anthracis could cover thousands of square miles and reach
millions of people with a single attack (as demonstrated by a series of pre-1969 US tests using simulants,
such as Operation Large Area Coverage).431,432 Although the biological properties of non-contagious
agents can facilitate their use in a weapon that can attack large areas, the ability of a non-contagious agent
to reach these large areas is highly dependent on the dispersal system, which require sophisticated
engineering skills to develop.433 Conversely, contagious agents could expose (and possibly infect) large
numbers of people over a wide area through the ongoing outbreak and the movement of infected people
without the need for a sophisticated dispersal device.
Insofar as an attack is desired to cause

disruption for a long period of time, B. Existing non-contagious agents are very good at causing
anthracis is also a good candidate because its mass casualties or contaminating a large area for a long
period of time if dispersed from a very sophisticated device.
spores can persist in buildings or in the soil
Contagious agents could have similar consequences with
for years. For instance, two and a half very simplistic dispersal methods.
months were required to perform
decontamination operations at the Landover
425 Centers for Disease Control and Prevention, “Clinical Signs and Symptoms of Influenza: Influenza Prevention & Control
Recommendations,” <http://www.cdc.gov/flu/professionals/acip/clinical.htm>;
Centers for Disease Control and Prevention, “Q Fever: Symptoms, Diagnosis, and Treatment.”
Biological Weapons Programme”.
Press
428 The Secretary of Defense, “Memorandum For the President, National Security Decision Memoranda 35 and 44,”
429 Jonathan B. Tucker, “Bioterrorism: Threats and Responses,” p. 286.
430 Sheila Van Cuyk et al., “Persistence of Bacillus thuringiensis subsp. kurstaki in Urban Environments following Spraying,”
Applied and Environmental Microbiology 77 (2011): p. 7954-7961,
431 Cole L, (1998) Clouds of Secrecy: The Army's Germ Warfare Tests Over Populated Areas Lanham: Rowman & Littlefield.
432 van Courtland Moon J, (2006) “The US Biological Weapons Program,” Deadly Cultures: Biological Weapons since 1945,
Cambridge: Harvard University Press.
433 For instance, the terrorist group Aum Shinrikyo tried to manufacture their own vehicular spray system, with poor result.
Richard Danzig et al., “Aum Shinrikyo: Insights into how terrorists develop biological and chemical weapons, second
edition,” Center for a New American Security, December 2012, p. 27,
http://www.cnas.org/files/documents/publications/CNAS_AumShinrikyo_SecondEdition_English.pdf.

mail facility in the aftermath of the 2001 anthrax attacks.434 Gruinard Island, the site of testing of anthrax
weapons in the former UK offensive weapons program, was dangerously contaminated for decades after
testing ceased.435 The entire topsoil of the island had to be decontaminated before the site was considered
safe for human occupation.436 Although the biological properties of B. anthracis enable a weaponeer to
deny access to a particular location for a long time after an attack, a population may simply avoid the
contaminated area and remain safe. In contrast, a contagious agent prolongs the effect of an attack
because the population itself carries the hazard, and, therefore, risk can be minimized only by reducing
human contact.
8.4.3.1 Contagiousness as a Desirable Characteristic to Increase the Footprint of an Attack
From the analysis above, the acquisition of a contagious agent would enable a weaponeer to expand the
footprint of their attack in terms of casualties, area and time. Moreover, the fact that an outbreak can
spread, sickening or killing victims beyond those infected by the initial attack, removes the requirement to
produce a sophisticated dispersal device to cause a mass casualty attack. For this reason, contagious
agents may be desirable by malicious actors who have significant skill in virology but no skill in
machining/engineering to make a munition.
However, contagious agents have drawbacks—primarily that their affect is difficult to predict or control.
Some state and sub-state groups may not desire an agent that could infect their soldiers or supportive
populations, or could cause mass fatalities. For instance, the United States’ now defunct offensive
biological weapons program considered a potential agent’s ability for human-to-human transmissibility as
a negative characteristic.437,438 This feature may be especially problematic if their supporters live
primarily in low-income countries because global outbreaks there may be more severe than in high-
income countries where the public health infrastructure is more robust. However, unlike the United
States, the Soviet Union’s program sought out highly contagious pathogens for at least some of the lethal
pathogens in its arsenal.439 At the state actor level, weaponeers may covertly stockpile a vaccine to
mitigate friendly losses to their contagious agent.
If a contagious, lethal agent is desired by a weaponeer, they could choose to work with a wild type agent
other than the influenza viruses or coronaviruses. Of the non-GoF Select Agents, smallpox virus, Yersinia
pestis and the filoviruses (Ebola and Marburg viruses) are the only viruses that have a high case fatality
rate and are significantly contagious. Y. pestis causes pneumonic plague if inhaled by a victim. This
pathogen is often described as being highly transmissible, with a historical R0 at or above that for
influenza strains.440,441,442 However, these historical studies draw upon past outbreaks in areas that do not
434 Dorothy A. Canter et al., “Remediation of Bacillus anthracis Contamination in the US Department of Justice Mail Facility,”
Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science 3, no. 2 (June 2005): p. 119-127,
http://online.liebertpub.com/doi/abs/10.1089/bsp.2005.3.119.
435 “Britain’s ‘Anthrax Island’,” BBC News, July 25, 2001, http://news.bbc.co.uk/2/hi/uk_news/scotland/1457035.stm.
436 Ibid.
437 US Department of the Army, “US Army Activity in the US Biological Warfare Programs, Volume 1,” February 24, 1977,
Unclassified, p. 50-51, http://nsarchive.gwu.edu/NSAEBB/NSAEBB58/RNCBW_USABWP.pdf.
Press
439 Ibid.
440 Gani R, Leach S, (2004) “Epidemiologic determinants for modeling pneumonic plague outbreaks,” Emerging Infectious
Diseases 10, no. 4 p.608-614, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3323083/
441 Nishiura H, et al., (2006) “Transmission potential of primary pneumonic plague: time inhomogeneous evaluation based on
historical documents of the transmission network,” Journal of Epidemiology and Community Health 60, no.7 p.640-645,
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2566243/
442 Coburn BJ, Wagner BG, Blower S, (2009) “Modeling influenza epidemics and pandemics: insights into the future of swine
flu (H1N1),” BMC Medicine http://www.biomedcentral.com/content/pdf/1741-7015-7-30.pdf

mirror the modern US.443,444 Indeed, most of the secondary cases of pneumonic plague described occurred
in household members of the ill, who also were the primary caregivers at a time when hospitalization of
the critically ill was still rare. Estimates of the R0 of pneumonic plague in the modern United States
suggest that an outbreak would extinguish rather rapidly.445 Similarly, outbreaks of Ebola virus disease in
Africa were associated with an R0 that approached that of influenza or SARS.446,447,448 However, much of
the transmission was in makeshift healthcare facilities suggesting that the R0 in the US would be much
smaller.449,450 Smallpox virus is held in just a few, highly secure locations throughout the world and,
therefore, may be difficult for an adversary to acquire. That being said, an adversary able to leverage the
information produced by GoF experiments may well be able to use non-GoF approaches to synthesize
smallpox virus de novo using well described rescue systems for other orthopoxviruses.451 Similarly,
although there is no environmental reservoir of SARS-CoV, the significant transmissibility and lethality
of the wild type strain may motivate a weaponeer to use reverse genetics to synthesize it or attempt to
acquire it from a laboratory.
If a contagious, incapacitating agent is desirable, not many options to acquire such an agent are available.
Wild type pandemic or seasonal influenza viruses are obviously highly contagious and incapacitating,
however residual immunity from past outbreaks may hamper the spread of these illnesses.452,453 An
adversary could choose a wild type strain that has not circulated in several decades to reduce the effect of
residual immunity in the population. Alternatively, modified strains of many pathogens, including mumps
virus and measles virus, which are
already highly contagious, could be A contagious pathogen may be desirable by a scientifically trained
made to overcome protective adversary with minimal engineering skill or by a state. Smallpox
immunity using techniques similar to virus and SARS-CoV are the only wild type pathogen that is
those used in GoF experiments. deadly and as transmissible (or more) than influenza virus. Wild
type influenza viruses are contagious, especially if a decades old
However, these experiments would be
strain is used to minimize the effect of residual immunity, and
associated with their own information highly incapacitating.
risk and, therefore, are not explored
further here.
8.4.3.2 Enhanced Growth as a Desirable Characteristic to Increase the Footprint of an Attack
Because pathogens are self-replicating, to increase the footprint of an attack, an adversary could simply
grow more of the agent. This goal could be accomplished by increasing the volume of culture used or by
443 Gani R, Leach S, (2004) “Epidemiologic determinants for modeling pneumonic plague outbreaks,” Emerging Infectious
Diseases 10, no. 4 p.608-614, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3323083/.
444 Hinckley AF, et al., (2012) “Transmission dynamics of primary pneumonic plague in the USA,” Epidemiology and Infection
140, no. 3 p. 554-560.
445 Ibid.
446 Adnan Khan et al., “Estimating the basic reproductive ratio for the Ebola outbreak in Liberia and Sierra Leone,” Infectious
Diseases of Poverty 4, no. 13 (February 2015), http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4347917/.
447 G. Chowell et al., “The basic reproductive number of Ebola and the effects of public health measures: the cases of Congo
and Uganda,” Journal of Theoretical Biology 229, no. 1 (July 2004): p.119-126.
448 Zhi-Qiang Xia et al., “Modeling the transmission dynamics of Ebola virus disease in Liberia,” Scientific Reports 5 no. 13857
(September 2015): p. 1-13, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4561958/pdf/srep13857.pdf.
449 Ibid.
450 see also: Joseph A. Lewnard et al., (2014) “Dynamics and control of Ebola virus transmission in Montserrado, Liberia: a
mathematical modelling analysis,” Lancet Infectious Diseases 14, no. 12 p. 1189-1195,
http://www.thelancet.com/pdfs/journals/laninf/PIIS1473-3099%2814%2970995-8.pdf
451 See FOUO addendum for examples.
452 For evidence of residual immunity, see for example: Pérez-Trallero E. (2009) “Residual Immunity in Older People Against
the Influenza A(H1N1) – Recent Experience in Northern Spain,” Eurosurveillance 14, no. 39
http://www.eurosurveillance.org/ViewArticle.aspx?ArticleId=19344.
453 On the effect of residual immunity on spread, see for example: Xu-Sheng Zhang et al., (2015) “Co-circulation of influenza
A virus strains and emergence of pandemic via reassortment: the role of cross-immunity,” Epidemics 5, no. 1 p. 20-33.

increasing the viral titer in the culture. This rationale supports the consideration of enhanced growth of a
pathogen in culture or eggs as a dual-use phenotype because an adversary can produce more pathogen
with the same amount of resources if the pathogen can grow to a high titer. We therefore evaluated how
producing more pathogen for a biological attack, in terms of the amount of pathogen released, relates to
the footprint of the attack, particularly in terms of number of victims infected (Data is provided in the
FOUO addendum). The results show that, for small amounts of pathogen effectively released, the number
of initial infections scales well with the amount of pathogen produced (that is, 10-fold more pathogen
released leads to 10-fold more initial infections). For larger amounts of pathogen, the increase in
casualties begins to taper off as the amount of pathogen released increases. Specifically, for every 10-fold
increase in pathogen released, the number of initial infections produced increases by only two- to three-
fold or less. Our results show that increasing the growth of a wild type strain will have a limited effect on
the amount of casualties produced even for actors who do not have access to industrial scale production
facilities. In short, even relatively poorly growing pathogenic strains of influenza virus grow well enough
such that producing more pathogen produces limiting returns for use in an attack.
Moreover, if a malicious actor wants to infect as many people as possible, spending the effort to modify a
strain to become high growth is probably not worth the risk that its transmissibility would be decreased.
By their nature, transmissible agents would increase the footprint of the attack, potentially by several
orders of magnitude if an ongoing and global outbreak can be sparked. An attack that initially infects
several thousand people will quickly grow out of local control (as demonstrated in the Risk Assessment
of Accidents and Natural Disasters) and, therefore, the number of casualties produced by the attack will
depend on the pathogenicity and transmissibility of the
pathogen in the context of a global epidemic and not the initial Experiments that enhance the growth
number infected. In Chapter 5, it is demonstrated that very few in culture of the GoF pathogens are
covert infections of the public are required to seed nearly of minimal information risk because
guarantee that an outbreak would escape local control for the producing more agent results in few
influenza viruses (10 or fewer), so the initial number of people additional casualties.
exposed could be very small indeed.
8.4.3.3 Countermeasure Resistance and Evasion of Existing Immunity as Desirable Characteristics to

Increase the Footprint of an Attack
The GoF phenotype of evasion to medical countermeasures includes the ability to evade the protection
afforded by vaccines, antivirals and diagnostics. The evasion of diagnostics is not particularly relevant to
GoF diseases released intentionally. SARS and MERS diagnostics are used in the US only for people
already thought to be infected with the disease based on their clinical symptoms.454 Since no specific
treatments for these diseases currently exists, these diagnostics would simply be used to direct public
health measures.455,456,457 Given the importance of quarantine and isolation in the control of coronavirus
454 Centers for Disease Control and Prevention, (2014) “Middle East Respiratory Syndrome (CDC) - CDC Laboratory Testing
for Middle East Respiratory Syndrome Coronavirus (MERS-CoV),” http://www.cdc.gov/coronavirus/mers/lab/lab-
testing.html.
455 During the 2003 SARS-CoV epidemic, Ribavirin was used; however, it “did not appear to have a significant effect,” and a
study of patients treated with Ribavirin indicated “that ribavirin provided no benefit in the resolution of symptoms or
survival.” In: Els Keyaerts, Leen Vijgen, Marc Van Ranst, “Current Status of Antiviral Severe Acute Respiratory Syndrome
Coronavirus Research,” Coronaviruses: Molecular and Cellular Biology, ed. Volker Thiel (Norfolk: Caister Academic
Press, 2007), p. 328.
456 Centers for Disease Control and Prevention, (2015) “Middle East Respiratory Syndrome (MERS)”
http://www.cdc.gov/coronavirus/mers/about/prevention.html.
457 World Health Organization, (2013) “Severe Acute Respiratory Syndrome (SARS)”.
http://www.who.int/immunization/topics/sars/en/

outbreaks, diagnostics could be used to direct public health resources to prevent the outbreak escaping
local control.458
Because pandemic and seasonal influenza viruses are highly contagious, the evasion of diagnostics would
have little consequence to the eventual extent of an outbreak, but may complicate the efficient use of
antivirals by confounding the identification of patients who are infected with influenza. For strains of
avian influenza modified to be contagious between humans, the evasion of a diagnostic would have just
as few benefits to a malicious actor as it would for the use of a coronavirus.
In an intentional attack with influenza virus, evasion of immunity induced by vaccination is of little
relevance because we presume the attack would be a surprise and the US would not have prepared
sufficient stocks of a protective vaccine ahead of time. For this reason, several months would pass before
the vaccine would be ready for deployment, at which time the disease would have spread globally (or, if
poorly contagious, extinguished by itself). However, the vaccine could still be used to limit the casualties
and temporal duration of the outbreak (see Figure 9.5 in Chapter 9 to see how the timing of the
deployment of a vaccine affects global casualties). However, a virus that can overcome protective
immunity induced by any vaccine would increase both the casualties of an attack and its duration.
Because no approved vaccines for the coronaviruses currently exist, malicious actors have no need to
make a strain of these viruses able overcome protective vaccination.
An actor wishing to leverage a recently circulating strain of influenza may desire to modify their strain to
avoid protective immunity. Residual immunity in the population can significantly reduce the chance that
an outbreak would escape local control and the consequences of an outbreak (Figures 6.35 and 6.57 in
Chapter 6). However, avoidance of residual immunity can be obtained either through GoF methods or by
the selection of a wild type strain that has not circulated recently. Infections by the SARS and MERS
coronaviruses are sufficiently rare that an adversary has no need to create a strain that can evade residual
immunity from a past infection.
Antiviral resistance of influenza viruses would be useful to an adversary to increase the casualties caused
by an outbreak in the United States and to increase the chance that a local outbreak escapes local control
(as shown in Figures 6.53 for seasonal influenza and 6.55 for pandemic influenza in Chapter 6). Given
that the majority of the world does not have access to the amount of antivirals that the United States does,
antiviral resistance has little influence on global consequences.
8.4.4 Gain of Function Strains Compared to Naturally Occurring Strains
8.4.4.1 A non-unique information risk inheres in experiments describing the creation of highly
transmissible, highly deadly strains of pathogens
When compared to strains of pathogens created via the use of GoF research information, we find that
naturally occurring strains generally are not both highly pathogenic and highly transmissible (with the
exception of wild type SARS-CoV). Although SARS-CoV has an R value greater than one, the fact that
the virus is transmissible only after symptoms present, the disease is largely spread within the medical
system, and the long incubation period of the disease makes outbreaks of SARS-CoV relatively easy to
control (as described in Chapter 4, past outbreaks have witnessed a two-fold drop in the R value of a
SARS outbreak after control measures are implemented). Only smallpox virus, which exists only in a few
laboratories, has a case fatality rate greater than 10% and is transmissible enough to cause a pandemic (an
458 See in particular figure 2 in: Simon Cauchemez et al., “Middle East respiratory syndrome coronavirus: quantification of the
extent of the epidemic, surveillance biases, and transmissibility,” Lancet Infectious Diseases 14, no. 1 (January 2014): p. 50-
56.

R value significantly greater than one in the context of a robust public health response). Because stocks of
smallpox virus are tightly controlled, an adversary might turn to GoF studies to acquire a pathogen that is
both highly pathogenic and highly transmissible. We note that no GoF study to date has conclusively
produced strains with the combination of the desired phenotypes because of weaknesses of animal models
in predicting pathogenesis in humans. Moreover, whether enhancement of a phenotype, like
transmissibility, would be sufficient to achieve the traits desirable by a weaponeer for an attack on a
human population remains unclear. That is, an experiment may show an increase in the transmissibility of
a pathogen among ferrets, but this observation cannot be translated into a specific R0 value in a human
population (or if the increase is sufficient to obtain the desired weapon characteristics). With these
limitations in mind, the following GoF results would be of concern:
• Seasonal/pandemic influenza virus that retains its transmissibility but is modified to have a case
fatality rate greater than 10%,
• Avian influenza viruses that are modified to be as transmissible as seasonal influenza but retain
their high fatality rate, and
• SARS/MERS-like CoV that is made more transmissible.
Scientific communications that detail the creation of strains with these traits would be of concern because
they would provide a route to the acquisition of a pathogen as useful to a weaponeer as smallpox virus.
Importantly, we have no data that speaks to the possibility that these phenotypes are achievable in the
laboratory. Perhaps, due to the epidemiology of influenza and the coronaviruses, high mortality (in excess
of that associated with the 1918 pandemic strain) and transmissibility are conflicting phenotypes because
the very ill do not contact many others during the contagious phase of the illness outside of a hospital.459
Moreover, these phenotypes will not emerge by chance in the laboratory. Any experiment that selects for
one phenotype is likely to allow other phenotypes to drift. That is, experiments that focus on enhancing
transmissibility alone are likely to arrive at viruses that are optimized for transmissibility. In those same
experiments, the viruses obtained can drift to less pathogenic forms because selection for this trait is not
maintained. In fact, this phenomenon was observed in the Fouchier experiment with H5N1, albeit not in
the Sutton 2014 experiment with H7N1. 460,461 In contrast, experiments that do not involve selection but
the systematic manipulation of the components of the virus, could lead to strains that have a variety of
phenotypes that are enhanced compared to a parental strain. For example, influenza reassortment studies
that choose one or two genomic segments from a highly transmissible influenza strain and switch them
for the cognate segments from a highly pathogenic strain can result in a strain where both phenotypes are
enhanced compared to the parents.462
Although some GoF studies could produce information of use to an adversary wishing to obtain highly
transmissible, pathogenic agents, these studies are not the only means of acquiring this type of agent.
Given that the sequence of smallpox virus is public, a technically sophisticated actor could use methods
similar to those employed in GoF laboratories to synthesize all (or the unique parts) of the smallpox
459 Interview comment by Dr. Ian Lipkin in: Donald G. McNeil Jr., “How a Mild Virus Might Turn Vicious,” The New York
Times, June 8, 2009, http://www.nytimes.com/2009/06/09/health/09flu.html?_r=0.
460 See Table 1 in: Sander Herft et al., “Airborne Transmission of Influenza A/H5N1 Virus Between Ferrets,” Science 336 no.
6088 (June 2012): p.1534-1541.
461 “The present findings show that adaptation of the H7N1 isolate does not appear to substantially decrease the virulence of the
virus.” In: Troy C. Sutton et al., “Airborne Transmission of Highly Pathogenic H7N1 Influenza Virus in Ferrets,” Journal of
Virology (2014): p. 6623-6635, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4054360/.
462 Ying Zhang et al., “H5N1 Hybrid Viruses Bearing 2009/H1N1 Virus Genes Transmit in Guinea Pigs by Respiratory
Droplets,” Science 340, no. 6139 (June 2013): p. 1459-1463, http://www.sciencemag.org/content/340/6139/1459.

genome and use it to rescue live virus.463 To support this effort, rescue systems are published already for
the orthopoxviruses.464
Moreover, researchers could use pathogens not listed in the GoF framework, but manipulate the same
traits considered in the framework to obtain highly transmissible and lethal strains of pathogens.
Examples include:
• Strains of filoviruses or henipaviruses that retain their pathogenicity but are modified to be much
more transmissible, and
• Strains of highly transmissible agents (like polio virus or measles virus—neither one of which is a
select agent) that have been modified to become more deadly and/or overcome protective
vaccination.
Finally, note that the reconstructed wild type strain of the 1918 influenza virus is predicted to be highly
transmissible and significantly pathogenic, and methods to recreate this virus through reverse genetics are
routine in influenza laboratories. Similarly, the wild-type strain of SARS-CoV is highly transmissible and
significantly pathogenic as well (as described in Chapter 6, further enhancing the transmissibility of
SARS-CoV increases the chance that a global outbreak would occur because of the susceptibility of
outbreaks caused by the wild type virus to control measures). Therefore, GoF studies may be of use only
if an adversary wishes to obtain a strain that is more pathogenic than that the 1918 pandemic influenza
strain or more transmissible than SARS-CoV.
8.4.4.2 A non-unique information risk inheres in the experiments describing the creation of highly
transmissible strains of influenza virus with specific enhancements in morbidity
Wild type influenza strains are well-suited for use as incapacitating agents because they have a small
mean infectious dose and cause a debilitating illness with a short incubation time. However, unlike other
pathogens researched in offensive weapons programs, influenza does not cause a particularly long lasting
or chronic illness. For this reason, a strain of influenza that produces a much more protracted course of
illness or chronic illness would provide an advantage over naturally occurring strains because no naturally
occurring strain with this combination of phenotypes is known to exist.
As above, GoF research is not a unique pathway to obtain this type of dual-use information. For example,
an adversary could produce strains of mumps, polio or measles that can overcome protective vaccination,
to obtain an agent with significant morbidity and even greater contagiousness than is currently possible
with influenza virus. Moreover, the experimental means to produce viral strains that can overcome
vaccination is well established. As such, this method of obtaining a highly-contagious, incapacitating
agent requires relatively little “research”. In contrast, the mechanisms underlying the nature of
pathogenicity of influenza, and what leads to a protracted illness, are still unknown. Moreover, no
research has pointed to any mechanism for chronic influenza infection and the creation of such a strain
would require a long-term research effort.
463 Institute of Medicine (US), Committee on the Assessment of Future Scientific Needs for Variola Virus, “Live Variola Virus:
Considerations for Continuing Research,” p. 13, 132.
464 An example of a publication of a rescue system for the orthopoxviruses is given in the FOUO appendix.

8.4.4.3 Evasion of Medical Countermeasures Is a Possible Future Information Risk
Evasion of medical countermeasures is a phenotype with minimal information risk today. Because no
medical countermeasures for the coronaviruses are in use today, there is no need to produce a strain of
these pathogens that can evade notional countermeasures. For the influenza viruses, vaccines, antivirals
and diagnostics are all in use. Because an influenza virus used in an attack would only serendipitously
match the seasonal vaccine produced and stockpiled by any nation, several months would pass between
the time an attack occurs and the time that a vaccine is even relevant to control the outbreak. After that
point, a vaccine has some utility for preventing mortality and morbidity and curtailing the outbreak (by
increasing the number of contacts needed to spread the disease). But, as shown in our Biosafety Risk
Assessment (Chapter 6) the evasion of vaccine-induced immunity at best provides a modest increase in
total mortality.
Similarly, when considering the mortality caused in a global outbreak, at best, antivirals reduce the death
rate of influenza by a few fold. However, in the US, because of our significant supplies of antivirals,
deaths can be greatly reduced by their effective administration. Conversely, then, an antiviral resistant
strain could increase the deaths from an attack with influenza virus in the US.
Diagnostics are currently used to direct the use of antivirals, which either are not plentiful enough
globally to influence consequences significantly or are so plentiful as to not be limiting in the US.
Diagnostics are also used to identify that an outbreak with a novel virus is occurring in the first place.
If an adversary wishes to use a wild type influenza strain in an attack, residual immunity from previous
outbreaks may limit the footprint of the attack. However, an adversary could simply use a wild type
influenza virus of a serotype that has not circulated recently to avoid this shortcoming. Once again,
published sequence information, combined with well-established protocols for the rescue of influenza
viruses, could be used to obtain these strains.
However, as vaccine technology advances, research on the evasion of medical countermeasures COULD
become an information risk. For example, once a universal influenza vaccine is developed, the evasion of
immunity induced by this vaccine may be critical for an adversary to cause an outbreak using influenza
virus as a weapon. Studies in other pathogens have described the development of strains able to overcome
protective vaccination due to the expression of exogenous genes (and not via escaping immune
recognition). Similarly, once systems are in place to develop a vaccine against a newly identified serotype
of influenza virus in a few weeks instead of a few months, evasion of induced immunity becomes more
useful to an adversary. Likewise, if antivirals become more widely available globally, research on the
evasion of that antiviral would pose an information risk. Clearly, as highly effective medical
countermeasures for the coronaviruses are developed, studies on their evasion would pose an information
risk. That being said, since these medical countermeasures do not yet exist, designing an experiment to
evade them is not currently possible. None of these technologies are likely to be deployed in the next five
years. However, once information is published, it is permanently available and retains its utility far into
the future. For this reason, the information risk relevant to the evasion of medical countermeasures should
be continually re-evaluated.
8.4.4.4 Enhanced Growth in Culture Affords Little Information Risk
As described in Sec. 8.4.3.2, strains with enhanced growth in vitro or in ovo can be used to produce more
agent to disperse from a weapon using the same amount of resources. However, even poorly growing
pathogenic strains can be grown in enough quantity with commonplace equipment to obviate the growth
of more to produce more casualties. In contrast, the use of contagious agents, by their nature, increases
the footprint (in terms of area affected and illnesses) by many orders of magnitude. For this reason,

publications on changes in this phenotype present very little information risk simply because the GoF
pathogens are contagious.
8.4.4.5 Summary of Possible Comparative Information Risk Arising from GoF Studies
Figure 8.2, below provides an overview of the potential information risk of GoF research.
Dual-Use GoF Phenotype Seasonal/Pandemic Avian Influenza Coronaviruses

Influenza
Enhanced transmissibility in mammals

Enhanced pathogenicity in mammals

Overcoming natural or induced
immunity
Evading diagnostics
Antiviral resistance
Enhanced production in cell culture or
eggs
Figure 8.2. Possible information risk arising from dual-use information relevant to GoF research. White
denotes that no significant information risk exists. Dark shading denotes a significant information risk, albeit
a risk that is not unique to GoF research. No information risks unique to GoF research were found.
8.5 Overview of the State of the Science of Dual Use GoF Information
In this section, we discuss how the existing body of GoF research already describes methods to obtain
strains of influenza- and coronaviruses with GoF traits, regardless of their true utility to a weaponeer. We
synthesize these two pieces of information to arrive at our final conclusions. That is, this section describes
the GoF information risk already realized through the previous publication of dual use information. To
maintain this discussion at the full-and-open level we have not cited the specific papers at issue and have
instead provided these in an appendix at the For Official Use Only level. Here, we simply characterize the
state of the science and describe the seminal publications.
We first describe the state of the science for research on influenza and later describe the state of the
science for the coronaviruses. We discuss each GoF phenotype in turn. Most scientific publications
investigate morbidity and mortality simultaneously, characterizing disease outcomes such as weight loss
or fever alongside death of the animals. For this reason, we discuss morbidity and mortality jointly as the
characteristic of pathogenicity.
8.5.1 State of the Science of GoF Experiments in Influenza Viruses
8.5.1.1 Enhanced transmissibility in mammals
Some research groups focus on understanding the factors that cause some H1N1 strains (such as the 2009
pandemic strain or 1918 pandemic strain) to be highly transmissible while other H1N1 strains (such as
avian strains and the Puerto Rico 8 strain) to not transmit in mammals. These groups use a variety of
methods to develop transmissible strains of H1N1 viruses, including:

• Simple reassortment, and
• Identification of mutations unique to transmissible strains followed by the use of reverse genetics
to introduce these changes in a non-transmissible strain.
These experiments, while demonstrating the ability to increase the transmissibility of influenza viruses,
are of limited dual utility because one of the parental strains already has the traits desired by a weaponeer
(transmissibility and pathogenicity). These experiments were executed to determine what factors are
sufficient to cause a poorly pathogenic virus into a highly pathogenic virus. An adversary gains no
advantage using one of these modified strains compared to the parental, pathogenic strain.
In contrast, in experiments dealing with avian-origin viruses, the parental strains are either highly lethal
OR highly transmissible in mammals and the manipulations described are required to obtain a strain that
is lethal AND transmissible in mammals. Several papers describe methods to manipulate highly
pathogenic strains of avian influenza from the H5, H7 or H9 subtype to arrive at a strain that is
transmissible by droplets in mammals (typically ferrets or guinea pigs). These studies use a variety of
methods including:
• Simple serial passaging in ferrets,

• Directed, sequential reassortment of an avian strain with a pandemic strain,
• Directed reassortment followed by serial passaging,
• Targeted mutagenesis followed by serial passaging, and
• Mutagenesis of HA, followed by selection based on its binding properties and the creation of a
chimeric virus.
Some of these experiments require only minimal skill in virology. The serial passaging experiment could
be repeated by an actor with limited skills in molecular biology to obtain a strain of avian influenza that is
transmissible in mammals. For actors with significant skill in molecular biology, these transmissible,
avian-origin influenza viruses could be synthesized using reverse genetics (a common practice in GoF
laboratories) because all investigators have published enough information on the specific molecular
changes observed in the avian-origin strains that are transmissible in mammals. Specifically, in the case
where serial passaging was used, the mutations acquired by the strains of interest were published. In the
case of the reassortment study, the gene segments that contributed to the transmissible strains from both
parental strain were published. For this reason, to leverage the information in these studies, an actor need
not repeat the experiments exactly, but simply reconstruct the viruses that the authors identified.
The abundance of literature describing the creation of avian-origin strains that are transmissible in
mammals indicates that the GoF information risk for enhanced transmissibility of influenza viruses is
already realized. Further publication on this topic would likely not exacerbate this risk because existing
published methods are relatively simple to replicate.
Any study, such as ours, that examines the existing literature on transmissibility of influenza viruses to
determine the dual-utility of the resulting information suffers from a critical shortcoming. The published
studies use too few animals in transmissibility experiments to understand exactly how transmissible these
newly developed strains are. That is, none of these studies compared the transmissibility of these strains
in a sufficient number of animals so that levels of transmissibility (compared to wild type seasonal or
pandemic influenza strains) could be determined. Moreover, ferrets in isolators do not interact with each
other the same way that people in a city do. For this reason, it is impossible to use laboratory experiments
to conclusively determine if the increase in transmissibility observed translates to a dangerous R0 value,
which is determined retrospectively by scrutinizing human epidemiological data. This distinction is

important because risk of an outbreak escaping local control and the risk of a resulting outbreak both
increase significantly as an influenza virus approaches a transmissibility comparable to that of a seasonal
influenza strain (see Chapter 6). Possibly all of these viruses may be somewhat transmissible, but in a
human population their R0 may be much less than one, which is the value required to cause a global
epidemic.
Experiments to compare the transmissibility of wild type strains of seasonal and pandemic strains to
modified strains would help determine if information risk inheres in further publications of strains that
have been modified to become transmissible. If
these modified strains are in fact transmissible, but Several groups have published simple methods
much less so than seasonal or pandemic strains, to increase the transmissibility in mammals of
pathogenic strains that were previously
then a remaining information risk would exist in
transmissible only in avians—therefore this type
further experiments that could identify a strain that of information risk is already realized
is as transmissible as seasonal or pandemic strains.
8.5.1.2 Enhanced pathogenicity
Several papers describe virulence factors necessary for the maintenance of pathogenicity in highly
pathogenic strains of influenza, such as the avian influenza and 1918 pandemic influenza viruses.
Typically, these researchers identify mutations unique to the highly pathogenic strains and introduce these
mutations into a less-pathogenic strain using reverse genetics. In a few experiments, researchers used
pathogenic strains that were isolated from patients with severe illness, attempting to determine what
characteristics made the strains infecting these patients even more pathogenic than the parental strain.
Other researchers use reassortment between highly pathogenic and non-pathogenic strains to obtain a
chimeric virus to identify the components of the pathogenic strain minimally necessary to confer
pathogenicity into the non-pathogenic background. These experiments, while demonstrating the ability to
increase the pathogenicity of influenza viruses, are of limited dual utility because one of the parental
strains already has the traits desired by a weaponeer (transmissibility and pathogenicity). If an adversary
has the pathogenic, transmissible strain, they gain no advantage from these manipulations.
In contrast, other experiments result in strains that are MORE pathogenic than any parental strain used. A
variety of methods are used to enhance the pathogenicity of influenza viruses, including:
• Random reassortment of seasonal and pandemic H1N1 strains,
• Directed, serial reassortment of an avian-origin strain with a pandemic strain,
• Site directed mutagenesis of residues associated with increased virulence in other strains to
increase the virulence of an avian strain, and
• Serial passaging in mice.
Although the serial passaging studies require minimal skill in molecular biology, they are also of
marginal dual-utility due to the limitations of the mouse model system of predicting pathogenicity of a
strain in humans (or ferrets).465 That said, these experiments indicate that an adversary seeking to develop
a more pathogenic virus could serially passage their strains in ferrets (or humans) to obtain a virus that is
more pathogenic. However, sometimes these serial passaging experiments lead to less pathogenic strains.
465 For example, Natalia A. Ilyushina et al., “Adaptation of Pandemic H1N1 Influenza Viruses in Mice,” Journal of Virology 84
no. 17 (September 2010): 8607-8616.

For actors with significant skill in molecular biology, strains with enhanced pathogenicity could be
synthesized using reverse genetics (a common practice in GoF laboratories) because all investigators have
published the specific molecular changes observed in the strains with enhanced pathogenicity. For this
reason, to leverage the information in these studies, an actor need not repeat the experiments exactly, but
simply reconstruct the viruses that they have identified.
The abundance of literature describing the

Several groups have published methods that require skills
creation of influenza virus strains with in molecular biology to increase the pathogenicity of
enhanced pathogenicity indicates that the influenza virus strains—therefore this type of information
GoF information risk for this trait is already risk is already realized. But a small remaining risk exists
realized. However, leveraging this from the publication of simple methods to the same result.
information requires skill in molecular Another modest risk inheres in the publication of methods
biology (specifically, reverse genetics), the to modify influenza viruses to cause chronic illness.
appropriate facilities and equipment. The
publication of simple selection methods supporting the conclusion that virulence could be increased in
relevant animal models via this simple method poses a remaining information risk. In contrast, we found
no publications describing the creation of an influenza virus strain that can produce prolonged or chronic
illness, which is important in the context of malicious actors seeking to produce an incapacitating agent.
8.5.1.3 Enhanced transmissibility while maintaining/enhancing pathogenicity
Experiments in this category are of particular concern because they could enable a hostile actor to obtain
a strain that has a combination of pathogenicity and transmissibility that surpasses all wild type, human
pathogens except for smallpox virus. Several papers describe methods to manipulate highly pathogenic
strains of avian influenza from the H5 or H7 subtype to arrive at a strain that is transmissible in mammals
(typically ferrets). Many of these studies also test the resulting strains for pathogenicity. The studies of
interest here use a variety of methods including:
• Simple serial passaging in ferrets, and

• Directed, sequential reassortment of an avian-origin strain with a pandemic strain.
In the serial passaging experiment, the researchers claim that no loss of pathogenicity is observed
compared to the highly pathogenic parental strain after ten passages in ferrets. However, too few animals
were used to assess pathogenicity to detect some loss of pathogenicity. In the reassortment study, the
authors show that two of the four transmissible strains they identified are more pathogenic than either
parental strain (although they tested this phenotype in mice). Although the authors make this claim, their
experimental design may not have been sufficient to detect true differences in pathogenicity from the
parental strain relevant to humans.
The serial passaging experiment could be repeated by an actor with limited skills in molecular biology to
obtain a strain of avian influenza that is newly transmissible in mammals and highly pathogenic (albeit
this actor would need animals and the facilities to contain them). Moreover, all investigators characterized
the resulting strains and published enough information (in the initial or follow-on papers) so that the
strain of the desired characteristics could be directly produced by reverse genetic methods. Specifically, in
the case where serial passaging was used, the mutations acquired by the strains of interest were published.
In the case of the reassortment study, the gene segments that contributed to the transmissible strains from
both parental strains were published.

For this reason, the GoF information risk
for enhanced transmissibility while Two groups have published simple methods to increase
maintaining pathogenicity of influenza the transmissibility in mammals of pathogenic strains of
viruses is already realized. Further influenza that were previously transmissible only in
publication on this topic would likely not avians while maintaining (or enhancing) pathogenicity.
exacerbate this risk because existing Therefore, this type of information risk is already realized.
published methods are relatively simple to
replicate. Actors skilled in reverse genetics could instead recreate the strains with the desired traits that
these groups characterized.
As discussed in the enhanced transmissibility section above, our retrospective study of the literature
cannot conclusively determine the dual utility of these publications. Further studies with more relevant
model animals would provide useful information about whether the resulting strains are likely to be both
more pathogenic and highly transmissible in humans. Also, our study (and any animal study) is unlikely
to conclusively determine if a transmissible strain would be similarly transmissible to seasonal influenza
in human populations.
8.5.1.4 Overcoming natural or induced immunity
As discussed above, the evasion of immunity is a desirable trait for a malicious actor if they wish to use a
strain of influenza that recently circulated (and, therefore, need to overcome the significant residual
immunity that exists in the population). For this reason, the information risk related to the ability of a
modified strain to overcome natural or induced immunity inheres in the methods to foster an antigenic
change in any desired virus strain. Therefore, methods published using attenuated strains are still relevant
to this information risk. All of the published papers reviewed in this study focus on elucidating the
mechanisms by which new strains with different antigenic profiles evolve. The methods involved include:
• Identification of unique changes between a parental strain and an antigenically distinct strain,
followed by the introduction of these unique changes into the parental strain by reverse genetics,
• Serial passaging in cells in the presence of neutralizing antibodies, and
• Serial passaging in immunized animals.
The serial passaging experiments could be repeated by an actor with limited skills in molecular biology to
obtain an antigenically distinct strain of influenza. Although one of the studies resulted in antigenically
distinct strains that are as pathogenic (or even more so, in mice) than the parental stains, other studies that
use large numbers of passages create significantly attenuated strains and still others do not characterize
the pathogenicity of their strains (however, fitness was assessed and maintained). Some studies using
reverse genetics produce strains that are antigenically distinct and pathogenic. No matter which method
was used, all investigators in the reviewed studies characterized the resulting strains and published
enough information (in the initial or follow-on papers) so that the strain with the desired characteristics
(distinct antigenicity and pathogenicity) could be directly produced by reverse genetic methods by a
malicious actor with significant skill in molecular biology.
For this reason, the GoF information risk for creating strains of influenza that are antigenically distinct
from their parental strain is already realized. Further publication on this topic would likely not exacerbate
this risk because existing published methods are relatively simple to replicate.

No papers were found that describe the
development of influenza strains able to Several groups have published simple methods to create
antigenically distinct strains of influenza virus.
overcome protective immunity without creating Therefore this type of information risk is already
an antigenically distinct virus. Methods have realized. To our knowledge, no one has published
been published describing the use of other methods to create strains of influenza virus able to
pathogens (viruses and bacteria) expressing overcome the protection afforded by vaccination in
exogenous factors to create strains that enable general, which poses a remaining information risk.
the pathogen to kill infected hosts despite being
effectively immunized with a vaccine matched
to the serotype of the pathogen. However, these pathogens have larger and more plastic genomes than the
influenza viruses and therefore, the development of such a strain of influenza would be a major research
undertaking. Moreover, the benefits of this type of research would be highly suspect. Regardless of the
reasons not to perform this research, the publication of methods to produce strains of influenza able to
overcome immunity regardless of its serotype would pose an information risk.
8.5.1.5 Evasion of diagnostics
Current generation influenza diagnostics function either via the recognition of epitopes in the virus (or
antibodies to these antigens generated by a patient) or via recognition of unique sequences in the genome
of the virus. Diagnostics that function by leveraging antibodies could be evaded in much the same way
that host immunity can be evaded, so this possibility will not be discussed further.
No papers reviewed in this study discussed the production of strains of influenza that can evade diagnosis
by the alteration of its genetic makeup (except for changes in the genome that lead to changes in
antigens). Actors with skills in molecular biology (and knowledge of the genetic targets of the assays)
could create strains of viruses with a series of silent mutations (mutations that alter the genomic material
but do not change the encoded proteins) to
confound recognition. However, codon usage Several groups have published simple methods to create
in viruses is sometimes tightly linked to antigenically distinct strains of influenza virus which can
fitness (or other desired traits) and, therefore, evade diagnostics that use the antigenic properties of the
a malicious actor must test their new strains to virus. Therefore, this type of information risk has already
ensure that all desired phenotypes were not been realized. No methods have been published
describing the modification of influenza strains to evade
lost. In any case, the publication of methods
diagnostic methods reliant on unique genomic
that demonstrate how to evade diagnostics via signatures. Therefore, this type of information risk
the alteration of the genome pose a remaining remains.
information risk.466
8.5.1.6 Antiviral Resistance
Because the creation of antimicrobial resistant strains is part of the drug-development process and part of
risk assessment process for determining when the effectiveness of antimicrobials may expire, several
publications on the creation of influenza strains resistant to antivirals could be found. The methods used
involved:
• Identification of unique changes between a parental strain and a drug resistant strain, followed by
the introduction of these unique changes into the parental strain by reverse genetics,
• Serial passaging in cells in the presence of low concentrations of the antiviral, and
466 Wong E, et al. (2010) “Codon usage bias and the evolution of influenza A viruses,” BMC Evolutionary Biology 10, 253

• Infection of animals treated with sub-optimal concentrations of the antiviral.
Serial passaging and in vivo experiments could be repeated by an actor with limited skills in molecular
biology to obtain an antiviral-resistant strain of influenza. In vivo methods also simultaneously select for
fitness (and in some cases, pathogenicity), suggesting that the resulting strains might still be useful in a
weapon. Studies using reverse genetics produce strains with several mutations, some of which lead to
antiviral resistance and others compensate for the defects in pathogenicity or fitness caused by the
primary mutation. These investigators characterized the resulting strains and published enough
information so that the strain with the desired characteristics (antiviral resistance and pathogenicity) could
be directly produced by reverse genetic methods by a malicious actor with significant skill in molecular
biology.
For this reason, the GoF information risk for creating Several groups have published simple
strains of influenza that are antiviral resistant is already methods to create antiviral resistant
realized. Further publication on this topic would likely strains of influenza. Therefore this type of
not exacerbate this risk because existing published information risk is already realized.
methods are relatively simple to replicate.
8.5.1.7 Increased production in cell culture or eggs
The vast majority of studies identified in which the production of viruses in cell culture or eggs was
enhanced involved the introduction of some components of pathogenic strains into attenuated strains.
These studies were designed to create attenuated strains with the immunoreactive antigens of the
pathogenic strains suitable for use as vaccines. Several studies discuss the generation of strains with
enhanced growth properties by reassorting pathogenic influenza viruses with attenuated strains adapted
for growth in eggs or culture.467 However, these strains were chosen because they simply express the HA
and NA antigens in a virus suitable for vaccination.
Others have adapted attenuated strains to achieve a 100-fold increase in titer after growth in cell culture
by serial passaging, but this method is likely to allow pathogenicity and transmissibility to drift.468 Others
have used random mutagenesis of the HA gene to identify high growth mutants, but these mutants are
likely to be poorly pathogenic.469
In contrast to these studies, other researchers have serendipitously found that some pathogenic strains
demonstrate increased growth in cells or eggs using a variety of methods, including:
• Serial passaging in animals, and

• Identification of unique changes between a parental strain and a highly pathogenic strain,
followed by the introduction of these unique changes into the parental strain by reverse genetics
The researchers leveraging reverse genetics introduced changes found in a highly pathogenic strain into a
less pathogenic background to determine if these changes were sufficient to increase pathogenicity. They
467 Zhang W et al. (2011) “Increase in viral yield in eggs and MDCK cells of reassortant H5N1 vaccine candidate viruses
caused by insertion of 38 amino acids into the NA stalk,” Vaccine 29, vol 45: 8032-41.
468 For example: Murakami S., Horimoto T,, Ito M,, Takano R,, Katsura H,, Shimojima M,, Kawaoka Y., “Enhanced growth of
influenza vaccine seed viruses in vero cells mediated by broadening the optimal pH range for virus membrane fusion,”
Journal of Virology 86 (2012): 1405-1410.
469 Ye J. et al. (2015) “Error-prone pcr-based mutagenesis strategy for rapidly generating high-yield influenza vaccine
candidates,” Virology 482: 234-243.

serendipitously, but perhaps not surprisingly, found that the growth of these pathogenic virus strains in
culture or in vivo was enhanced. Since a malicious actor would likely desire this combination of
phenotypes, this type of study generates biosecurity information risk. However, beyond using reverse
genetics to synthesize these strains, these pathogenic, high-titer strains could be directly acquired from
clinical samples.
Actors unable to acquire the specific

Several groups have published approaches of developing highly
highly pathogenic strains studied and pathogenic strains of influenza that grow to a higher titer than
who lack skills in molecular biology their parental strains. Repeating these experiments requires
skill could instead repeat one of the skills in molecular biology. Those without molecular biology
serial passaging studies published. skills could repeat a serial passaging experiment, but they must
However, both of these studies used take measures to retain pathogenicity of the final strain. For
mice as their model system to identify these reasons, little information risk remains for this trait.
strains that were more pathogenic and
grew to a higher titer in cell culture or in eggs. Because of the weakness of the mouse animal model,
passaging in mice may not lead to strains that are pathogenic in people. Presumably, passaging the virus
in another animal model would retain the virulence of the strain (as shown in the passaging experiments
above) and may result in higher growth variants.
8.5.2 State of the Science of GoF Experiments in the Coronaviruses
8.5.2.1 Enhanced transmissibility in mammals
No model system currently exists for the study of transmissibility of SARS- or MERS-CoV and,
therefore, no studies have described methods to increase the transmissibility of these viruses. Therefore a
significant information risk remains for any studies that describe the development of an animal model of
transmission. However, these studies are necessary to understand the evolution of the viruses, their life
cycle, their associated pathology, and pathways for developing vaccines and drugs.
8.5.2.2 Enhanced pathogenicity
SARS- and MERS-CoV do not normally infect mice, so the virus must be manipulated to infect this host
to study pathogenicity. In fact, mouse-adapted SARS-CoV cannot effectively bind to or infect human
cells. For this reason, although several groups described strains of coronaviruses that have enhanced
pathogenicity compared to the parental, mouse-adapted coronavirus strains, these viruses are presumably
acting in mice more like the wild type SARS- and MERS-CoV do in humans. These experiments are
performed to learn how SARS- and MERS-CoV became pathogenic to humans and not to determine how
they could become more pathogenic in the future.
Model animal systems for the study of SARS- or
If new experimental systems were developed, an
MERS-CoV can give us information on how these
information risk would be possible. That being pathogens became dangerous to humans but
said, even if these experiments were to be probably not how they can become more
conducted, very little room for the enhancement of dangerous. If new experimental systems were
pathogenicity is possible due to the high case- developed, an information risk would be possible.
fatality rate and severity of the symptoms of the
diseases these pathogens cause.
8.5.2.3 Enhanced transmissibility while maintaining/enhancing pathogenicity
Currently, no model system for the study of transmissibility of SARS- or MERS-CoV exists and,
therefore, no studies have described methods to increase the transmissibility of these viruses. Therefore a
significant information risk remains.

8.5.2.4 Evading Diagnostics
Current generation coronavirus diagnostics function either via the recognition of epitopes in the virus (or
antibodies generated against these antigens by a patient) or via recognition of unique sequences in the
genome of the virus. Diagnostics that function by leveraging antibodies could be evaded in much the
same way that host immunity can be evaded. We found two papers that describe the selection of a SARS-
CoV that can escape binding by antibodies. In these papers, researchers serially passaged the virus in cells
in the presence of antibodies derived from infected patients. In one study, the researchers studied how
pathogenic the escape mutants were and found that some retained their pathogenicity.
No papers reviewed described the production of strains of coronaviruses that can evade diagnosis by the
alteration of its genetic makeup (except for changes in the genome that lead to changes in antigens).
Actors with skills in molecular biology (and knowledge of the targets of the assays) could create strains of
viruses with a series of silent mutations (mutations that alter the genomic material but do not change the
encoded proteins) to confound recognition.
However, codon usage in viruses is We found two groups that have published simple methods
sometimes tightly linked to fitness (or other to create antigenically distinct strains of SARS-CoV which
desired traits).470 For this reason, a malicious can evade diagnostics that use the antigenic properties of
actor must test their new strains to ensure the virus. Therefore, this type of information risk has
that the desired phenotypes were not lost. already been realized. No methods have been published
The publication of methods that demonstrate describing the modification of coronavirus strains to evade
how to evade diagnostics via the alteration of diagnostic methods that target genomic sequence.
the viral genome pose a remaining Therefore, this type of information risk remains.
information risk.
8.5.2.5 Overcoming countermeasures and immunity
As no approved vaccines or antivirals exist for prevention or treatment of infections caused by the
coronaviruses, experiments of this type are not possible so they pose no information risk. Moreover,
methods to evade antibody-based countermeasures and small molecule countermeasures are well
established for other pathogens, including influenza. If such countermeasures were developed in the
future, an adversary is likely able to leverage any of these methods to develop strains of the coronaviruses
that are resistant to the countermeasures.
Moreover, infections by the SARS and MERS coronaviruses are sufficiently rare so that an adversary has
no need to create a strain that can evade residual immunity from past infections.
8.5.2.6 Increased production in cell culture or eggs
We found no papers describing the enhancement of viral growth in culture or eggs. However, SARS- and
MERS-CoV grow to a relatively high titer in culture, suggesting enhancement of this trait is unnecessary.
8.5.3 Overview of the State of the Science of GoF Experiments
Figure 8.3, below summarizes the analysis of the literature already published relevant to GoF research.
We found that much of the dual-use information that could arise from GoF experiments has already been
published for the influenza viruses. For this reason, much of the information risk for this pathogen has
already been realized. The remaining information risk inheres in the creation of simpler experimental
470 Gu W. et al. (2004) “Analysis of synonymous codon usage in SARS Coronavirus and other viruses in the Nidovirales,”
Virus Research 101, vol 2: 155-161.

approaches to the development of strains with enhanced pathogenicity or enhanced growth, and any
method that leads to the creation of strains that can avoid protection by any vaccine or the evasion of
diagnostics via alteration of its genome. In contrast, the lack of model animal systems for the study of
transmission or enhanced pathogenicity of the coronaviruses leaves a significant information risk if these
systems were developed. That is, future experiments that describe how to make a strain of the
coronaviruses more contagious (or more deadly) have a significant information risk. Although medical
countermeasures for the coronaviruses do not yet exist, if they were developed a malicious actor could
easily leverage simple experimental procedures published for other pathogens to create strains that
overcome the countermeasures.
Dual-Use GoF Phenotype Influenza Coronaviruses
Enhanced transmissibility in
mammals
Enhanced pathogenicity in Published methods require skills in

mammals molecular biology. No publications
exist on creation of influenza
strains that lead to chronic illness.

Overcoming natural or induced Via the creation of antigenically N/A

immunity distinct strains only
Evading diagnostics Evasion of immunological Evasion of immunological

diagnostics only diagnostics only
Enhanced production in cell culture Published methods require skills in N/A

or eggs molecular biology.
Figure 8.3. Status of the publication of potentially dual-use information relevant to GoF research. White
denotes that no significant information risk is left either because the relevant information has already been
published or the resulting trait is not dual-use. Dark shading denotes a lack of publications on the topic so a
significant information risk may remain. Grey denotes that some information risk may be remaining.
8.6 Evaluation of the Capability and Intent of Malicious Actors to Leverage Dual Use
Information
Given the analysis above, non-unique information risk resides in the ability to create strains of highly-
transmissible, anti-viral resistant pathogens that have a high case fatality rate and are not limited by
residual immunity brought about by the natural circulation of similar strains. That being said, this risk is
already realized because the methods to produce these strains is already published. For malicious actors
interested in developing a uniquely capable incapacitating agent, research on the development of
influenza strains that produce chronic or long-lasting illness would be of interest.
We acknowledge that GoF research is just one means through which a malicious actor may obtain such a
pathogen. By leveraging GoF information, sophisticated actors could use reverse genetics to create a
strain that was previously described by other researchers with all desired characteristics. Actors with
limited skills in molecular biology could use the selection experiments described in the literature to attain
strains that may have all the desired traits, but extensive testing would be necessary to identify a strain

with all such characteristics. This section describes the actors with the capability and motivation to seek
to leverage this information.
State actors, who in the past have sought deadly strains and incapacitating strains of pathogens for use in
offensive weapons programs, clearly often have the ability to acquire the equipment and expertise to use
reverse genetics to create any strain of influenza or coronavirus described in the literature. Moreover,
states likely would have the ability to design and produce a cache of vaccines (in the case of modified
influenza viruses) that could protect their own population from the contagion that may spread from their
intended target (and to protect their workers during the development and weaponization process). For this
reason, if states were to leverage GoF information for malicious use, they would likely target information
on the influenza viruses (instead of the coronaviruses) and would be uninterested in strains that can avoid
any type of vaccine protection. That being said, secrecy inside a state program may hamper the
coordination of the offensive and defensive components of a biological weapons program. For example,
those working on countermeasures in the Soviet biological weapons program lacked the clearance to
know about the offensive side.471,472 Moreover, the stockpiling of vaccines specific to a strain of
influenza virus that has no risk of a natural outbreak could undesirably broadcast information about a
state’s offensive program.
Other malicious actors, from individuals to terrorist groups, have so far shown very little ability to acquire
or grow biological agents. The few terrorist groups that have attempted to do so have relied on bacterial
agents, not viral agents. The publically available literature suggests that no sub-state group has so far
demonstrated the scientific sophistication and the resources necessary to leverage GoF information for
malicious purposes.
In theory, a lone outsider (or small group of outsiders) with scientific training may have the ability to
perform the manipulations necessary to obtain modified pathogens via relatively simple methods.
Alternatively, if unconstrained by time and resources these technically trained actors may be able to use
reverse genetics to obtain any desired strain. Industry standards for customer and sequence screening, in
part supported by U.S. government guidance, may prevent actors from acquiring synthesized pathogens
via genetic material synthesized by industry. However, some gene synthesis (domestically and
internationally) companies do not voluntarily follow the industry standards or U.S. government guidance.
Lone scientific actors working in GoF laboratories (i.e. insiders) can simply directly acquire the desired
strains as described in Chapter 7, above. Given the access controls implemented in containment
laboratory, there is little opportunity for an insider to carry out a clandestine development and testing
program inside the laboratory they legitimately have access to. Therefore, insiders are not considered a
driver of information risk because they have access to strains and unpublished data.
Press.
472 Ouagrham-Gormleym S (2014) Barriers to Bioweapons Ithaca: Cornell University Press.

A malicious actor developing an engineered strain would
Of the potential malicious actors, only
presumably need the resources to test the newly state actors have the resources, technical
developed strain to ensure that it has all the traits desired. sophistication and motivation to leverage
Animal testing facilities are expensive and difficult to dual-use information arising from GoF
covertly obtain, establish and operate without running the studies. These states could protect their
risk of self-infection and/or exposing the public. Sub- own population by secretly stockpiling an
state groups, however, are likely to be satisfied with effective vaccine against their modified
more rudimentary animal testing than those conducted by agent, suggesting they may prefer to
scientists seeking to publish in a peer-reviewed target influenza viruses over the
publication.473 coronaviruses.
If somehow a scientific actor can create a facility that is sufficiently remote to develop the strains and
perform the needed testing without being noticed (by intelligence gathering or by causing an outbreak),
they could produce a highly contagious, highly lethal strain of virus. These properties avoid an often
noted shortcoming of small groups wishing to produce an effective biological weapon: that they lack
either the scientific expertise to create a useful biological agent or the engineering expertise to create a
useful biological munition (both of which are normally needed to create a weapon capable of inflicting
mass casualties). Infections caused by rudimentary devices (like a spray bottle) may be enough to seed a
global pandemic causing mass casualties. However, as noted in Chapter 7, above, we know of no lone
actor who has ever desired to cause a global pandemic (albeit, a small group, R.I.S.E, did). Moreover,
terrorist organizations and individuals appear to act more on the emotional and societal motivations
towards radicalization and violence, suggesting a greater focus on the immediate, rather than long-term,
response or outcome of an act.474 Of course, historical examples do exist of individuals and terrorist
groups that invested significant resources and time in planning of an attack (see Sec. 7). Whether the
calculus of long-term illness or mass casualty through infection is attractive to sub-state actors is not
known from publicly available literature.
One final caveat: the unsophisticated actor today could leverage advancing technologies to gain a
significant body of skills and knowledge. Software is being developed to automate the research process,
with the possible outcomes of increasing reproducibility and decreasing human involvement in the
experimental process.475 Finally, several online blogs, websites, video journals, and analytic technologies
increasingly are being used to identify experimental protocols, troubleshoot experimental problems, and
design experimental reagents such as DNA primers for polymerase chain reaction experiments.476
Together, these changes may increase democratization of life sciences experiments and lower the level of
skill and advanced scientific knowledge needed to conduct experimental procedures.477
473 Danzig R, et. al. (2012) “Aum Shinrikyo: Insights into how terrorists develop biological and chemical weapons, second
edition,” Center for a New American Security
474 De Angelis T (2009) “Understanding Terrorism,” American Psychological Association.
http://www.apa.org/monitor/2009/11/terrorism.aspx. Accessed on October 27, 2015.
475 For example, see Amyris Genome Compiler. Accessible at http://www.genomecompiler.com/amyris-dna-construction-on-
genome-
compiler/?utm_source=refferal_website&utm_medium=press_release&utm_term=5_8_2015&utm_content=website&utm_c
ampaign=amyris_alpha_program. Accessed on September 12, 2015.
476 For example, see JOVE. Accessible at http://www.jove.com/. Accessed on September 12, 2015.
477 For additional discussion, J Revill and C Jefferson. Tacit knowledge and the biological weapons regime. Sci Pub Policy.
2013; KMVogel. Phantom Menace or Looming Danger? A New Framework for Assessing Bioweapons Threats. 2013. The
Johns Hopkins University Press. (Baltimore, Maryland),

9 Benefit Assessment of GoF Research
Overview of Results 243
9.1.2 Influenza viruses 244
9.1.3 Summary 245
Methodology 255
9.2.1 Purpose of This Task 255
9.2.2 Conceptual Approach to the Identification of Potential Benefits of GoF Research 255
9.2.3 Characterizing the expected scientific information and products derived from GoF studies 257
9.2.4 Identifying potential benefits of GoF research to scientific knowledge, public health,
and medicine 257
9.2.5 Identifying current practices in medical countermeasure development and production
that rely on GoF approaches 259
9.2.6 Identifying gaps in scientific knowledge about PPPs and gaps in public health and
medical capabilities related to the prevention and control of PPP outbreaks 259
9.2.7 Crosswalking GoF research outcomes to the gaps in scientific knowledge, public health,
and medicine 259
9.2.8 Assessing the barriers to the realization of GoF and alt-GoF studies 260
9.2.9 Assessing if alternate experimental approaches and other scientific innovations could
lead to the same benefits 261
9.2.10 Evaluating the Globalization Potential of GoF Benefits 262
9.2.11 Quantitative Analysis of GoF Benefits 263
Coronaviruses: Benefits of GoF research 263

9.3.1 Summary 263
9.3.2 Overview of the GoF research landscape involving coronaviruses 268
9.3.3 Identification of potential benefits and limitations of GoF research involving CoVs 270
9.3.4 Identification of alt-GoF that provide similar potential benefits to the GoF being examined 276
9.3.5 Comparison and analysis of the potential benefits of GoF approaches versus alt-GoF
approaches 282
Introduction to GoF Research Involving Influenza Viruses 286

9.4.1 Overview of the landscape of GoF Research Involving Influenza Viruses 286
9.4.2 Use of attenuated strains of influenza viruses 287
9.4.3 Organization of the assessment of the benefits of GoF research involving influenza viruses 288
Influenza viruses: Benefits of GoF Research that Enhances Virus Production 289
9.5.1 Summary 289
9.5.2 Overview of GoF research landscape: enhanced virus production 290
9.5.3 Identification of potential benefits and limitations of GoF approaches 292
9.5.4 Identification of the potential benefits and limitations of alt-GoF that provide similar
potential benefits to the GoF being examined 295
approaches 300
Influenza viruses: Benefits of GoF research that enhances mammalian adaptation and
transmissibility 301
9.6.1 Summary 301

9.6.2 Overview of GoF research landscape: enhanced infectivity and transmissibility
in representative animal models 304
9.6.3 Identification of the potential benefits and limitations of GoF approaches 305
9.6.4 Identification of the potential benefits and limitations of alt-GoF approaches
that provide similar potential benefits to the GoF approaches being examined 317
9.6.5 Comparison and analysis of the potential benefits of GoF approaches versus
alt-GoF approaches 324
Influenza viruses: Benefits of GoF research that enhances virulence 329

9.7.1 Summary 329
9.7.2 Overview of GoF research landscape: enhanced pathogenicity in representative
animal models 332
that provide similar potential benefits to the GoF approaches being examined 340
Influenza viruses – Benefits of GoF research that leads to Evasion of Existing Natural or Induced
Adaptive Immunity 351
9.8.1 Summary 351
9.8.2 Overview of GoF research landscape: evasion of existing natural or induced adaptive
immunity 353
9.8.4 Identification of the potential benefits and limitations of alt-GoF approaches 362
approaches 368
Influenza viruses: Benefits of GoF Research that Leads to Evasion of Vaccines 371
9.9.1 Summary 371
9.9.2 Overview of GoF Research Landscape: Evasion of Vaccines 371
9.9.4 Identification of the potential benefits and limitations of alt-GoF approaches that
provide similar potential benefits to the GoF approaches being examined 373
approaches 373
Influenza Viruses: Benefits of GoF Research that leads to Evasion of Therapeutics 373
9.10.1 Summary 373
9.10.2 Overview of GoF research landscape: evasion of therapeutics 376
Influenza viruses: Benefits of GoF Research Involving Reassortment 393

9.11.1 Summary 393
9.11.2 Overview of GoF research that involves reassortment 394

Evaluation of the quantitative benefits of GoF research 403

9.12.1 Overview of GoF and alt-GoF benefits subject to quantitative analysis 404
9.12.2 Overview of GoF benefits not subject to quantitative analysis 405
9.12.3 Benefit Associated With Seasonal Influenza Vaccine 405
9.12.4 Benefit Associated with Pandemic Influenza Vaccine 406
Likelihood of GoF Strains Arising in Nature 408

9.13.1 Summary 408
9.13.2 Introduction 409
9.13.3 Evasion of existing natural or induced immunity (Antigenic Drift) 409
9.13.4 Evasion of Therapeutics 410
9.13.5 Enhanced Pathogenicity 411
9.13.6 Mammalian Adaptation and Enhanced Transmission in Representative Animal Models 413
Evaluation of the Globalization Potential of GoF Research 423

9.14.1 Summary of Findings 423
9.14.3 Potential Benefit 1- Improvements in the design and production of vaccines 425
9.14.4 Potential Benefit 2- Assistance in the development of new influenza or coronavirus
antivirals 432
9.14.5 Potential Benefit 3- Benefits to pandemic preparedness planning 439

9.1 Overview of Results
The Benefit Assessment evaluated the potential benefits of GoF experimental approaches involving
coronaviruses (CoVs) and influenza viruses to scientific knowledge and public health, including benefits
to biosurveillance, to the development of medical countermeasures (MCMs), and to decision-making in
public health policy. In each case, the ability of GoF approaches to address gaps in scientific knowledge
or shortcomings in public health was compared to the ability of alternative approaches to address those
same gaps, which enabled identification of the unique benefits of GoF research. Two types of alt-GoF
approaches were considered: alternative experimental approaches that can provide the same or similar
information and alternative scientific or technical innovations that can address the same public health
gaps through completely different mechanisms. Of note, unlike the risk assessment, the benefit
assessment was limited to the evaluation of GoF approaches that have been described in the scientific
literature.
Within the field of CoV research, GoF approaches in the following phenotypic categories were identified:
enhanced pathogen production, altered host range, enhanced virulence, and evasion of therapeutics in
development. Within the field of influenza research, GoF approaches in the following phenotypic
categories were identified: enhanced virus production, mammalian adaptation and enhanced
transmissibility, enhanced virulence, evasion of vaccines or therapeutics, and evasion of existing natural
or induced adaptive immunity. The following figure summarizes the results of the benefit assessment.
GoF Phenotype Coronaviruses Seasonal Influenza Animal Pathogenic

Viruses Influenza Reassortant
Viruses Influenza
Viruses*
Adaptation to mammals
Enhanced transmissibility N/A
Enhanced pathogenicity
Evasion of vaccines in N/A
development
Evasion of existing natural N/A N/A N/A
or induced adaptive
immunity
Evasion of therapeutics
Enhanced virus production
Reassortment (multiple GoF N/A
phenotypes possible)
*Pathogenic reassortants influenza viruses include reassortants comprised of gene segments from seasonal and
pandemic or seasonal and animal influenza viruses.
Figure 9.1 Summary of the benefits of GoF research by phenotype. White indicates that the phenotypic
change cannot be achieved or is not relevant (given the current state of model systems, the current state of
MCMs, or the biological characteristics of the virus). Dark grey indicates that the current phenotypic change
may be achievable but has not been undertaken in the scientific literature. Light grey indicates that the
approach provides unique benefits to scientific knowledge and/or public health. Medium grey indicates that
alternative approaches can address the same scientific knowledge or public health gaps that GoF approaches
can address. Note that medium grey does not indicate that GoF and alt-GoF approaches are equally capable
of addressing those gaps, simply that a more nuanced evaluation is needed to understand the relative value of
GoF and alt-GoF approaches.

The brief section that follows provides an overview of the GoF benefits identified in each phenotypic
category.
9.1.1 Coronaviruses
GoF approaches that alter host range and enhance virulence uniquely enable the development of animal
model systems that recapitulate human disease pathogenesis, which are critical for establishing the safety
and efficacy of candidate vaccines and therapeutics and for the study of CoV pathogenesis. This
manipulation to a new host typically attenuates virulence in the original host (in the case of SARS and
MERS-CoV, humans). GoF approaches that enhance virulence are also uniquely capable of
demonstrating that live attenuated vaccines (LAVs) do not recover virulence upon growth in vivo, an
important aspect of safety testing of candidate LAVs. Of note, this particular approach simply increases
the human health risk of the attenuated strain to approach that of wild type strains. GoF approaches that
enhance virulence represent the most efficient and effective strategy for discovering novel virulence
factors, which may be good targets for new therapeutics. However, several alternative strategies for the
development of new therapeutics are being actively pursued and have also shown promise. GoF
approaches that lead to evasion of therapeutics are critical for the development and regulatory approval of
new therapeutics. Because these therapeutics are not yet widely available, no increase in human health
risk is posed by resistant strains. GoF approaches that alter host range and enhance virulence provide
unique benefits to study cross-species adaptation and pathogenicity, but alternative approaches may also
be used.
9.1.2 Influenza viruses
Across all GoF phenotypes, GoF approaches provide unique benefits to the study of the mechanistic basis
of the phenotype under study as well as the evolutionary mechanisms driving acquisition of that trait,
though alternative approaches may also be used. Alternative approaches have stringent limitations for the
study of mechanisms underlying mammalian transmissibility of animal influenza viruses, as animal flu
viruses that efficiently transmit in humans do not exist in nature.
GoF approaches that enhance virus production are uniquely critical for the current ability to produce
sufficient and timely influenza vaccines for seasonal flu epidemics and flu pandemics and represent the
only strategy for improving existing vaccine production capabilities in the near-term. Of note, these
particular approaches attenuate an otherwise pathogenic strain while enhancing its growth properties.
GoF approaches that enhance the infectivity, transmissibility, and virulence of animal flu viruses generate
information that is used to evaluate the pandemic risk posed by circulating animal influenza viruses
detected through surveillance. In conjunction with several other types of data, including epidemiological
data and virological data, this information informs pandemic risk assessments of circulating influenza
viruses, which guide downstream decision-making about investments in pre-pandemic vaccine
development and other pandemic preparedness initiatives. In general, GoF data moderately contribute to
the overall risk associated with a particular virus, relative to other types of data that are considered.
However, GoF data play an important role in rapid risk assessments when novel flu viruses first emerge in
human populations due to the early availability of sequence data. These risk assessments facilitate more
rapid initiation of response activities such as pre-pandemic vaccine development. In addition, GoF data
may guide the selection of a particular virus to use as the basis of a pre-pandemic vaccine, among a set of
strain candidates that otherwise have similar epidemiologic and phenotypic properties. Of note,
realization of these benefits is subject to significant advancements in the state of knowledge about

mechanisms underlying mammalian adaptation, transmissibility, and virulence, as well as expansion of
sequencing capabilities across public health laboratories involved in influenza surveillance.
GoF approaches that enhance the infectivity and virulence of influenza viruses are also used to develop
animal models that support the study of disease pathogenesis and medical countermeasure (MCM)
development. GoF approaches that lead to evasion of therapeutics are critical for the development and
regulatory approval of new therapeutics. Because these therapeutics are not yet widely available, no
increase in human health risk is posed by resistant strains. However, similar approaches using licensed
therapeutics inform therapeutic recommendations for seasonal influenza infections and pandemic
preparedness initiatives for high-risk animal influenza viruses, but phenotypic approaches for antiviral
sensitivity testing are also used for these purposes. GoF approaches that lead to evasion of vaccines are
uniquely capable of determining whether viruses can acquire mutations to escape neutralization of
candidate broad-spectrum or universal influenza vaccines, a critical aspect of testing the potential field
efficacy of vaccines in development. Because these vaccines are not yet widely available, no increase in
human health risk is posed by strains that can overcome the protection afforded by these vaccines. GoF
approaches that lead to evasion of existing natural or induced immunity have potential to aid antigenic
surveillance of seasonal influenza viruses and to improve the efficacy of seasonal influenza vaccines,
though alternative approaches for achieving these goals are also being pursued. Notably, this GoF benefit
is subject to advancements in the state of knowledge about the mechanistic basis of antigenic drift as well
as expansion of sequencing capabilities across public health laboratories involved in global influenza
surveillance. Finally, GoF studies involving reassortment, which may lead to one or more phenotypic
changes, are uniquely capable of providing information that can be used to prioritize community-level
interventions aiming to prevent opportunities for co-infections that could lead to the generation of
reassortant viruses with phenotypic properties of concern.
9.1.3 Summary
Chapter 9 provides a relatively brief description of all of the benefits of GoF research that were identified
in this study. Chapter 16 provides a fully referenced and in-depth discussion of these findings and
includes a summary table for each GoF benefit, which describes the relative strengths and limitations of
GoF and alt-GoF approaches that can achieve that benefit. As the relative ability of a given GoF (or alt-
GoF) approach to address a particular scientific knowledge or public health gap often hinges on nuanced
differences between the benefits and limitations of different approaches, readers who seek an in-depth
understanding of the benefits of GoF research are directed to chapter 16.
The following table summarizes the set of benefits identified for each GoF phenotype and directs readers
to the relevant sections and summary tables that accompany each benefit.

Table 9.1. List of potential benefits of GoF research involving coronaviruses.
Benefit Potential Benefit Unique Sections Crosswalk Time Horizon for Time Horizon - Notes
Category Benefit Table(s) Benefit
Realization
Develop in vitro model
Scientific
systems for studying bat No 9.3.5.1.1; 15.1.4.1 15.3 Immediate
Knowledge
CoVs
Altered host range (mammalian adaptation)
Gain insight into
Scientific
mechanistic basis of cross- No 9.3.5.1.1; 15.1.3.1 15.1 Immediate
Knowledge
species adaptation
Develop in vitro model
Scientific
systems for studying bat No 9.3.5.1.1; 15.1.4.1 15.3 Immediate
Knowledge
CoVs
Scientific Develop animal models for
No 9.3.5.1.3; 15.1.4.2 15.4 Immediate
Knowledge studying CoV pathogenesis
Vaccines and Develop animal models for Development and licensing of new
No 9.3.5.4.1; 15.1.4.2 15.4 Long term
therapeutics testing candidate MCMs MCMs is a long process
Develop model system for
Vaccines and Development and licensing of new
testing broad-spectrum No 9.3.5.4.2; 15.1.5.3 15.8 Long term
therapeutics MCMs is a long process
efficacy of MCMs
Enhanced virulence
Gain insight into
Scientific
mechanistic basis of CoV No 9.3.5.1.2; 15.1.3.2 15.2 Immediate
Knowledge
virulence
Scientific Develop animal models for
No 9.3.5.1.3; 15.1.4.2 15.4 Immediate
Knowledge studying CoV pathogenesis
Vaccines and Develop animal models for Development and licensing of new
No 9.3.5.4.1; 15.1.4.2 15.4 Long term
therapeutics testing candidate MCMs MCMs is a long process
Identify new therapeutic Development and licensing of new
Therapeutics No 9.3.5.3.1; 15.1.5.2 15.6 Long term
targets therapeutics is a long process

Table 9.1. List of potential benefits of GoF research involving coronaviruses.
Benefit Potential Benefit Unique Sections Crosswalk Time Horizon for Time Horizon - Notes
Category Benefit Table(s) Benefit
Realization
Support development of Development and licensing of new
Vaccines Yes 9.3.5.2; 15.1.5.1.1 15.5 Long term
live attenuated vaccines vaccines is a long process
Evasion of therapeutics in development
Gain insight into
9.3.5.3.2; Development and licensing of new
Therapeutics therapeutic’s mechanism of No 15.7 Long term
15.1.5.2.2. therapeutics is a long process
action
Facilitate regulatory
9.3.5.3.2; Development and licensing of new
Therapeutics approval of new Yes N/A Long term
15.1.5.2.3 therapeutics is a long process
therapeutics
Inform development of
therapeutic strategies that 9.3.5.3.3; Development and licensing of new
Therapeutics Yes N/A Long term
minimize development of 15.1.5.2.4 therapeutics is a long process
resistance

Table 9.2. List of potential benefits of GoF research involving influenza viruses.
Agent Benefit Potential Benefit Unique Sections Crosswalk Time Horizon Time Horizon -
Category Benefit Table(s) for Benefit Notes
Realization
Gain insight into the
Scientific
Seasonal, animal mechanistic basis of high No 9.5.5.1; 15.2.4.2 15.10 Immediate
Knowledge
growth of vaccine viruses
Status quo – GoF
Enable the sufficient and approaches are
Seasonal, animal Vaccines timely production of flu Yes 9.5.5.2.1; 15.2.3 15.9 Immediate currently a key
vaccines aspect of flu
vaccine production
GoF insights can
9.5.5.2.2; 15.11; be applied to
Shorten future vaccine
Seasonal, animal Vaccines No 9.5.5.2.3; 15.12; Near term vaccine production
production timelines
15.2.4.3 15.13 without the need
for FDA approval
Mammalian adaptation and enhanced transmissibility
Gain insight into mechanistic 15.14;
Scientific basis of mammalian 15.15;
Animal No 15.3.3 Immediate
Knowledge adaptation and acquisition of 15.16;
transmissibility 15.17
Inform surveillance of Information from
circulating animal flu viruses GoF studies can be
Animal Surveillance by enabling sequence-based No 9.6.5.2; 15.3.4 15.18 Near to long term immediately
prediction of adaptation and applied to
transmissibility surveillance and
Inform pandemic risk downstream
assessment of animal flu decision-making
viruses and downstream about pandemic
Animal Policy No 9.6.5.3; 15.3.5.2 15.19 Near to long term preparedness
decision-making about
investments in pandemic activities, including
preparedness activities pre-pandemic

Realization
vaccine
development.
However, this
benefit is
constrained by
scientific
Guide selection of strains for uncertainties
Policy, 9.6.5.3; Near term to long associated with the
Animal pre-pandemic vaccine Yes 15.4
Vaccines 15.3.5.2.5.1 term data. The benefit of
development
GoF information to
surveillance is
expected to
increase over time,
as the state of the
science advances.
Enhanced virulence
15.20;
Scientific Gain insight into mechanistic 9.7.5.1; 15.4.3.1;
Seasonal, animal No 15.21; Immediate
Knowledge basis of influenza virulence 15.4.3.2
15.22
Develop animal models for
Scientific 9.7.5.1.3;
Seasonal, animal studying influenza No 15.23 Immediate
Knowledge 15.4.3.3
pathogenesis
Inform surveillance of Information from
circulating animal flu viruses GoF studies can be
Animal Surveillance No 9.6.5.2; 15.3.4 15.18 Near to long term
by enabling sequence-based immediately
prediction of virulence applied to
Inform pandemic risk surveillance and
assessment of animal flu downstream
viruses and downstream decision-making
Animal Policy No 9.6.5.3; 15.3.5.2 15.19 Near to long term about pandemic
investments in pandemic preparedness
preparedness activities activities, including

Realization
pre-pandemic
vaccine
development.
However, this
benefit is
constrained by
scientific
Guide selection of strains for uncertainties
Policy, 9.6.5.3;
Animal pre-pandemic vaccine Yes 15.4 Near to long term associated with the
Vaccines 15.3.5.2.5.1
development data. The benefit of
GoF information to
surveillance is
expected to
increase over time,
as the state of the
science advances.
Development and
Support development of live 9.7.5.3.1; licensing of new
Seasonal, animal Vaccines Yes N/A Long term
attenuated vaccines 15.4.5.1 vaccines is a long
process
Improve safety of vaccine FDA approval may
production process by be needed for
9.7.5.3.2;
Animal Vaccines identifying virulence markers No 15.24 Intermediate term application of GoF
15.4.5.2
that can be removed from insights to vaccine
vaccine viruses production
Development and
Identify new therapeutic licensing of new
Seasonal, animal Therapeutics No 9.7.5.4; 15.4.5.3 15.25 Long term
targets therapeutics is a
long process

Realization
Development and
Develop animal models for licensing of new
Seasonal, animal MCMs No 9.7.5.5; 15.4.3.3 15.23 Long term
testing candidate MCMs MCMs is a long
process
Evasion of existing natural or induced immunity
15.26;
Scientific Gain insight into mechanistic
Seasonal No 9.8.5.1; 15.5.3 15.27; Immediate
Knowledge basis of antigenic drift
15.28
Improve antigenic surveillance Information from
by enabling sequence-based GoF studies can be
Seasonal Surveillance No 9.8.5.2; 15.5.4 15.29 Near to long term
prediction of antigenic immediately
phenotype applied to
surveillance and
downstream strain
selection for
seasonal flu
vaccines, but that
benefit is
constrained by
Increase the efficacy of scientific
seasonal flu vaccines by 9.8.5.3.1; uncertainties
Seasonal Vaccines No 15.30 Near to long term
improving strain selection 15.5.5.1 associated with the
capabilities data. The benefit of
GoF information to
surveillance is
expected to
increase over time,
as the state of the
science advances.

Realization
The development
of a universal or
Inform the development of broad-spectrum
9.8.5.3.2;
Seasonal, animal Vaccines universal or broad-spectrum No 15.31 Long term vaccine represents
15.5.5.2
vaccines a very scientifically
challenging
prospect
Evasion of Vaccines
The development
Test whether viruses can of a universal or
escape protective immunity broad-spectrum
Seasonal, animal Vaccines conferred by candidate Yes 9.9.5; 15.3.3 N/A Long term vaccine represents
universal or broad-spectrum a very scientifically
vaccines challenging
prospect
Evasion of therapeutics
Scientific Gain insight into mechanistic 15.32;
Seasonal, animal No 9.10.5.1; 15.7.3 Immediate
Knowledge basis of antiviral resistance 15.33
Improve surveillance for Information from
antiviral resistance by GoF studies can be
Seasonal, animal Surveillance No 9.10.5.2; 15.7.4 15.34 Near to long term
enabling sequence-based immediately
prediction of resistance applied to
Inform therapeutic surveillance and
9.10.5.3; downstream policy
Seasonal Policy recommendations for seasonal No N/A Near to long term
15.7.5.1 decisions, but that
flu

Realization
benefit is
constrained by
scientific
Inform pandemic risk uncertainties
assessment of animal flu associated with the
viruses and downstream data. The benefit of
Animal Policy No 9.10.5.3;15.7.5.2 N/A Near to long term GoF information to
investments in pandemic surveillance is
preparedness activities expected to
increase over time,
as the state of the
science advances.
Improve safety of vaccine FDA approval may
production process by be needed for
Seasonal, animal Vaccines identifying resistance markers No 9.10.5.4; 15.7.6 15.35 Intermediate term application of GoF
that can be removed from insights to vaccine
vaccine viruses production
Development and
Inform development of new 9.10.5.5.1; licensing of new
Seasonal, animal Therapeutics Yes N/A Long term
therapeutics 15.7.7.1 therapeutics is a
long process
Development and
Gain insight into therapeutic’s 9.10.5.5.2; licensing of new
Seasonal, animal Therapeutic No 15.36 Long term
mechanism of action 15.7.7.2.1 therapeutics is a
long process
Development and
Facilitate regulatory approval 9.10.5.5.2; licensing of new
Seasonal, animal Therapeutic Yes N/A Long term
of new therapeutics 15.7.7.2.2 therapeutics is a
long process

Realization
Inform development of Development and
therapeutic strategies that 9.10.5.5.3; licensing of new
Seasonal, animal Therapeutic Yes N/A Long term
minimize development of 15.7.7.3 therapeutics is a
resistance long process
Reassortment studies
Gain insight into mechanisms
Scientific
Seasonal, animal driving and underlying No 9.11.5.1; 15.8.3 15.37 Immediate
Knowledge
reassortment
Surveillance for
reassortant viruses
Inform assessment of the risks
is poor and must be
Seasonal, animal Surveillance posed by reassortant viruses No 9.11.5.2; 15.8.4 15.38 Long term
improved for
detected through surveillance
realization of this
benefit.
Inform pandemic risk Surveillance for
assessment of animal flu reassortant viruses
viruses and downstream 9.11.5.3.2; is poor and must be
Seasonal, animal Policy No 15.19 Long term
decision-making about 15.8.5.2 improved for
investments in pandemic realization of this
preparedness activities benefit.
Inform prioritization of
interventions that aim to
9.11.5.3.1;
Seasonal, animal Policy prevent the emergence of Yes N/A Near term
15.8.5.1
novel reassortant viruses in
human populations

9.2 Methodology
9.2.1 Purpose of This Task
The purpose of the qualitative benefit assessment (BA) is to provide information regarding the potential
benefits of GoF research to scientific knowledge, public health, and medicine, including benefits to
biosurveillance, decision-making in public health policy, and the development of vaccines, therapeutics,
and diagnostics. (Throughout the report, the term “public health” is used to encompass all applied benefits
to public health and medicine.) Economic benefits were not explicitly evaluated but are noted where
relevant. Similarly to the risk assessments, the benefit assessment will be comparative; that is, the benefits
of GoF studies are evaluated relative to the benefits of alternative experimental approaches that can
provide similar information or other scientific and technical innovations that can provide similar benefits.
In addition, the BA will seek to provide information regarding barriers to the realization of the benefits
and the global distribution of the benefits, two key considerations when weighing the potential benefits
against research risks that may be global and immediate.
9.2.2 Conceptual Approach to the Identification of Potential Benefits of GoF Research
The approach to the benefit assessment is founded on the concept that the benefits of scientific research
derive from applications of new scientific information or products to unanswered scientific questions or
unmet needs in public health and medicine (collectively referred to as “gaps”). To that end, a multi-step
process was used to identify the benefits of GoF research relative to alternative approaches, as illustrated
in Figure 9.2. First, a foundation for the analysis of benefits was established by independently (a)
characterizing the expected scientific information and products derived from GoF studies of potential
concern, and (b) identifying gaps in scientific knowledge about PPPs and gaps in public health and
medical capabilities related to the prevention and control of PPP outbreaks. Second, the scientific
information/products derived from GoF research were mapped (“crosswalked”) to the gaps in scientific
knowledge and public health. That is, for each scientific outcome of GoF research, the gaps in scientific
knowledge and public health that the information/product could address were identified; subsequently, the
mechanism by which the information/production could overcome shortcomings in that gap area was
determined. This crosswalk analysis was guided by the proposed benefits of GoF research, as suggested
by infectious disease researchers and “translators” involved in application of research to public health
challenges. The outputs of the crosswalk analysis–GoF research applications and their downstream effects
on the health of human populations–represent the potential benefits of GoF research. Third, alternate
experimental approaches and/or other scientific or technical innovations that could lead to the same or
similar benefits were identified. Fourth, the barriers to the realization of GoF and alt-GoF benefits were
assessed, including factors that impede the translation of the research as well as “downstream” factors that
limit its ultimate impact on human morbidity and mortality. Comparative analysis of the benefits afforded
by GoF research versus alternative approaches, in light of the barriers to the realization of each approach,
yielded insight into the unique benefits of GoF research. Fifth (not shown in Figure 9.2), the globalization
potential of GoF benefits that were found to be uniquely beneficial were analyzed. Lastly, the impact of
GoF benefits to the production of influenza vaccines on the public health burden of seasonal flu
epidemics and flu pandemics was quantitatively analyzed.

Figure 9.2. Conceptual approach and workflow for benefit assessment. “Relevant” scientific outcomes/products of alt-GoF research are those outcomes
that can address similar gaps as GoF research. “Relevant” gaps in scientific knowledge and public health are those gaps that can be addressed by GoF
research. Strategies for identifying the relevant outcomes and gaps will be described in detail below. Because the applications of new scientific
knowledge to public health are long-term and unanticipated (e.g. whether a newly discovered virulence factor is a good therapeutic target), the barriers
to realization of these long-term benefits will not be evaluated.

9.2.3 Characterizing the expected scientific information and products derived from GoF studies
The scientific body of work that falls within the definition of GoF research on PPPs was analyzed, as
informed by the NSABB’s Framework for Conducting Risk and Benefit Assessments of Gain-of-
Function Research and the USG funding moratorium on certain types of GoF research. Specifically, this
analysis included scientific research involving seasonal influenza viruses, pandemic influenza viruses
(e.g. 1918 pandemic influenza virus), swine influenza viruses, and avian influenza viruses, as well as
research involving SARS coronavirus, MERS coronavirus, and SARS/MERS-like bat coronaviruses.
Within each field of research, all experimental approaches that are reasonably anticipated to confer one or
more of the following phenotypic changes were evaluated:
• Enhanced pathogen production as a result of changes in the viral replication cycle or growth,
• Enhanced transmission in mammals, including altered host or tissue range and more efficient
transmission by contact or airborne routes,
• Evasion of existing natural or induced immunity, and
• Evasion of vaccines, therapeutics or diagnostics.
Subsequently, within each GoF phenotype and for influenza viruses and coronaviruses separately, a set of
general experiments that capture the range of GoF studies conducted in the published literature was
defined, termed the “landscape” of GoF research. Each general experiment is described by:
• Experimental goal(s) (e.g. gain insight into mechanisms of airborne transmissibility of influenza
viruses),
• Experimental approach (e.g. serial passaging of influenza virus in ferrets with selection for
airborne transmission),
• Virus strains that are used (e.g. animal-origin influenza strains), and
• Expected research output(s), including new scientific information and/or products (e.g. gain
insight into molecular mechanisms of airborne transmissibility of influenza viruses between
mammals and identify genetic determinants of airborne transmissibility in influenza viruses).
The list of expected scientific outcomes/products of GoF research of potential concern served as the
inputs of our crosswalk analysis. Specifically, scientific outcomes/products were mapped to gaps in
scientific knowledge, public health, and medicine in order to assess their potential benefits to science and
society.
9.2.4 Identifying potential benefits of GoF research to scientific knowledge, public health, and
medicine
Specific proposed benefits of GoF research to scientific knowledge and public health were identified, as
described by scientific researchers, including those who conduct GoF studies of potential concern on

influenza, SARS, and MERS, those who employ alternative approaches to study influenza, SARS, and
MERS, and other scientists from the public health, agricultural, and wildlife research communities, as
well as “translators” involved in applying GoF research to public health and medicine. This included
benefits in all benefit areas defined in the NSABB Framework for Conducting Risk and Benefit
Assessments of Gain-of-Function Research (i.e. scientific knowledge, biosurveillance, medical
countermeasures, decision-making in public health policy, and economic benefits), as well as any other
benefits suggested by GoF stakeholders. Several data sources informed this task. First, GoF stakeholders
have discussed potential benefits of GoF research during public meetings about GoF research and through
perspectives and research articles published in scientific journals. This data set was reinforced and
expanded upon via interviews with GoF stakeholders involved in conducting and translating scientific
research.
In total, 86 stakeholders from a variety of sectors were interviewed, as shown in Figure 9.3, below.
Benefit Analysis: SME Interviews
11
4
Government
8 Industry
Non-PPP research
PPP research
63
Figure 9.3. A pie graph showing the sector from which the 86 interviewees were drawn. This chart includes
senior research staff, postdoctoral fellows, and graduate students we interviewed during our site visits to labs
that conduct PPP research. “Translators” include government and industry personnel, as well as some PPP
researchers who are involved in translation activities, such as WHO strain selection meetings for the seasonal
influenza vaccine. Of note, several government personnel are also actively involved in PPP research.
As is described below, in addition to data about the proposed benefits of GoF research, interviews
provided data to support every subsequent task in the BA. Finally, additional data on potential
applications of GoF research was gathered through review of relevant scientific literature on infectious
disease surveillance, MCM development, and public health policy.
Taken together, this information enabled the development of a list of proposed benefits of GoF research
to scientific knowledge and public health, which informed two aspects of the subsequent analyses. First,
given these data, public health areas (e.g. pandemic risk assessment using surveillance data, development
of influenza vaccines, etc.) that encompass all potential GoF benefits described by scientists and
translators were defined, which were subjected to a gap analysis as described below. Second, the list of
proposed benefits of GoF research guided the crosswalk of the outputs of GoF research to gaps in
scientific knowledge and public health.

9.2.5 Identifying current practices in medical countermeasure development and production that
rely on GoF approaches
Whether and how GoF approaches contribute to current practices in the development and production of
influenza virus and coronavirus MCMs was determined. First, FDA regulations related to the approval of
MCMs were analyzed to determine whether GoF studies facilitate or are essential for any aspects of the
process, including analysis of whether resistance studies are required for the approval of new therapeutics
or vaccines, the role of the Animal Rule in the demonstration of MCM safety and efficacy, and other
relevant regulations. Second, the role of GoF approaches in current processes for egg- and cell-based
vaccine production was reviewed through analysis of the academic literature and through interviews with
industry and government personnel with expertise in influenza vaccine production. The continued
application of GoF research to these areas represents one type of potential benefit of GoF research.
9.2.6 Identifying gaps in scientific knowledge about PPPs and gaps in public health and medical
capabilities related to the prevention and control of PPP outbreaks
This task involved identification of gaps in scientific knowledge about PPPs and gaps in public health and
medical capabilities related to the prevention and control of PPP outbreaks that could potentially be
addressed by insights gleaned from GoF research. This analysis was undertaken for several reasons. First,
identification of alternative approaches that aim to address the same or similar gaps as GoF studies
requires a complete and nuanced understanding of the gaps and their role in the overall public health
process, as alternative approaches may benefit the same ultimate public health gap (e.g. delayed
availability of vaccines during an influenza pandemic) by addressing different shortcomings in the
process (e.g. increasing the rate of vaccine production versus developing pre-pandemic vaccines that can
be rapidly deployed during a pandemic). Second, this gap analysis enabled identification of scientific and
non-scientific barriers to the realization of the benefits.
Many gaps in public health and medicine cannot be addressed by biomedical research. Broadly speaking,
the scope of this analysis was bounded by the list of GoF benefit areas defined in the task above,
including biosurveillance, development and production of vaccines and therapeutics, and decision-making
for public health preparedness. Within each benefit area, the list of proposed benefits was further utilized
to focus on identifying and researching gaps that could be targeted by GoF research. Importantly, gaps
were evaluated independently of their relationship to GoF research. To understand critical gaps in
scientific knowledge about PPPs, the state of the science regarding how influenza viruses, SARS-CoV,
and MERS-CoV are transmitted between hosts, cause disease, overcome protective immunity, and evolve
new phenotypic characteristics was reviewed. Interviews with researchers and “translators,” as well as an
analysis of the scientific literature, provided information about gaps in public health and medicine.
Notably, this research attempted to identify not only the gaps that could be addressed by GoF studies, but
also who may use the outputs of GoF studies to address the gaps, so that these stakeholders could be
interviewed to validate the assessment of the benefits of GoF research (described below).
9.2.7 Crosswalking GoF research outcomes to the gaps in scientific knowledge, public health, and
medicine
The next phase of the analysis determined how the research outputs of GoF studies can address gaps in
scientific knowledge, public health, and medicine. This “crosswalk” was guided by the proposed benefits
of GoF research. For each potential application of GoF research, limitations of the experimental approach
that could influence the nature and scope of the benefit were considered. For example, a set of mutations
that confer efficient transmissibility to one strain of zoonotic influenza, identified through a GoF
experiment, may not lead to the same phenotypic changes in a different genetic context, which

compromises the utility of this information in aiding analysis of biosurveillance data. Additionally, the
analyses of public health systems, described above, were leveraged to understand the process by which
the immediate applications of GoF research ultimately reduce human morbidity and mortality caused by
influenza viruses and coronaviruses. That is, all the steps needed to realize the benefits from discovery
(e.g. identification of genetic markers that confer high growth to vaccine viruses) to immediate
application (e.g. incorporation of markers into vaccine viruses used for manufacturing, in order to shorten
production timelines by increasing the rate of viral antigen production) to ultimate impact on public
health (e.g. faster vaccine availability during a pandemic reduces morbidity and saves lives) were
delineated. For benefits related to the production of influenza vaccines, the effects of improving the
availability vaccines on human morbidity and mortality during outbreaks were further evaluated using
quantitative methods, as described in the “Quantitative Analysis of GoF Benefits” section below.
Collectively, the outputs of the crosswalk analysis – GoF research applications and their downstream
impacts on the health of human populations – represent the potential benefits of GoF research. Notably,
realization of some benefits may depend on other scientific and non-scientific factors, the implications of
which are explored in our assessment of barriers to the benefits, described below.
The concerns of researchers who have been critical of proposed benefits of GoF research, particularly
those concerned about whether and when benefits will be realized, were assessed for their relevance and
validity as part of our analysis of GoF research benefits. Similarly, concerns of public health, medical,
and public health policy professionals who explicitly critique proposed benefits of GoF research were
analyzed. Also, GoF stakeholders were re-engaged as needed to validate the assessment of the benefits of
GoF research.
9.2.8 Assessing the barriers to the realization of GoF and alt-GoF studies
One of the most challenging aspects of weighing the risks and benefits of GoF research is that there is a
temporal mismatch between the risks and the benefits of the research–the risks are assumed at the time
the research is conducted, while the benefits to public health and medicine may accrue in the future. To
enable the comparison of risks and benefits, the benefit assessment is structured to provide data about the
probability and likelihood that the potential benefits of GoF research will be realized.
To accomplish this goal, benefits to scientific knowledge and benefits to public health/medicine were
considered separately. Scientific insights have immediate intrinsic value and may also inform the
development of novel vaccines or therapeutics, surveillance strategies, and other advancements in public
health/medicine in the future. Because the nature and timing of such applications are difficult to predict
with certainty, this report acknowledges but does not attempt to elucidate or evaluate the unforeseen
applications of basic science research to public health or medicine for this analysis.
In contrast, the potential benefits of GoF research to public health/medicine involve clear applications of
scientific information gleaned through GoF studies to unmet needs in public health. However, unlike the
risks, which pose possible direct threats to humans, animals and the environment , the benefits involve
“upstream” aspects of the public health process (e.g. biosurveillance), and evaluating how and when the
benefits will improve the health of human populations is complex. That is, translation of the research may
depend on other scientific, technical, and regulatory factors (e.g. the need to gain FDA approval in order
to market a new therapeutic), and gaps or inefficiencies in downstream aspects of the public health
process (e.g. limited funding for investment in the development of pre-pandemic vaccines) may limit the
ultimate impact of the research application on human health. Collectively, these factors function as
“barriers” that reduce the likelihood and delay the timing of the realization of the benefits, although
significant uncertainties in when and whether barriers can be overcome preclude a meaningful
quantitative estimate of either parameter.

Instead, the gap analysis of public health capabilities was leveraged to elucidate the barriers to the
realization of each potential benefit and to determine the type of resources or advancements that are
required to overcome or circumvent each barrier. These advancements include advances in science and
technology, including additional research needed to strengthen the certainty of scientific conclusions (e.g.
whether a particular genetic marker for mammalian adaptation is conserved across strains), investments in
public health infrastructure and MCM development and changes in public health policies or regulations.
This analysis was informed by the concerns related to benefit realization that were identified in the
literature and through interviews with stakeholders. For all aspects of this task, scientists, public health
practitioners, MCM developers, public health policy-makers, and other GoF stakeholders previously
interviewed were re-engaged as needed to clarify opinions regarding GoF benefits and benefit concerns,
challenges in biosurveillance, MCM development, public health policy-making, and other topics.
9.2.9 Assessing if alternate experimental approaches and other scientific innovations could lead to
the same benefits
GoF studies comprise a subset of all research activities involving PPPs, and some alternative approaches
may pose less risk than GoF studies but yield the same or similar benefits. Two types of “alt-GoF”
approaches were considered. First, alternative experimental approaches that can address the same
scientific questions as GoF approaches were identified, for example loss-of-function versus gain-of-
function approaches for identifying determinants of pathogenicity. The second type of alt-GoF approach
considered is other scientific and technical approaches that can address the same public health gaps that
GoF can address, but using a completely different strategy. For example, GoF studies that increase the
yields of influenza vaccine viruses in eggs or cell culture may benefit influenza vaccine production by
shortening the time needed to produce the same number of vaccine doses. However, a completely
different strategy, the production of recombinant influenza vaccines using insect cells may also address
issues related to the timeliness and amount of vaccine available even though this alternate approach
shares no experimental features with the GoF approach. After considering these alt-GoF approaches, the
benefit assessment can identify those types of GoF studies that may provide unique benefits to scientific
knowledge and public health, which will complement the analysis of the net risks associated with the
conduct of GoF research relative to alternative approaches.
Alternative approaches that may yield similar information or public health impacts as GoF research were
identified by drawing on the alternative approaches suggested by infectious disease researchers and
translators during public meetings about GoF research (such as NSABB meetings and NAS symposia), in
perspectives published in scientific journals, and during interviews, as well as the scientific literature on
PPPs. Importantly, alt-GoF research spans a wide range of topics, and those alt-GoF studies that yield
information outside the scope of GoF research are not relevant for the analysis. For this reason, to focus
the analysis on those approaches that may inform the same or similar gaps as GoF research, alt-GoF
approaches were identified by starting with the set of scientific knowledge gaps that are targeted by GoF
studies and referencing the scientific literature to identify alt-GoF approaches that target those same gaps.
The analysis of alternative approaches that target similar public health gaps critically leveraged the
analysis of public health systems, in particular the understanding of how the steps from discovery to
application of GoF research participate in an overall system.
Subsequently, the potential benefits of alt-GoF approaches were identified through the same process as
for GoF studies: a crosswalk of the research outputs of alt-GoF studies or the products of alternative
scientific/technical innovations to gaps in scientific knowledge and public health that can be addressed by
GoF research. Similarly, the barriers to the realization of alt-GoF benefits were assessed through

identification of co-factors needed for the translation and downstream public health impacts of alt-GoF
approaches.
Ultimately, the goal of the benefit assessment is to identify the benefits of GoF research of concern
relative to alternative experimental approaches that may pose less risk, as described below. A list of
benefits was compiled, as well as the scientific and non-scientific co-factors required for realization of
each benefit for each GoF research approach of potential concern. To provide a comparison, a similar list
was compiled for each alt-GoF approach evaluated. Evaluation of the unique benefits involved
comparison of GoF and alt-GoF benefits, in light of barriers to realization of each set of benefits. To
identify the unique benefits of GoF research to scientific knowledge, the benefits of GoF research and
those of alternative experimental approaches were compared. Identification of the unique benefits of GoF
research to public health involved additional comparison of the benefits of GoF research to those of
alternative scientific and technical innovations that address the same public health gap through different
mechanisms. Beyond an explicit consideration of barriers, a variety of factors were considered when
comparing the benefits of GoF research to alt-GoF research, including the ability of an approach to:
• Provide causative vs correlative (associative) data,
• Provide direct evidence of a phenomenon vs indirect evidence (e.g. showing that pathology
changes by manipulating the virus vs manipulating the host of a virus),
• Provide the ability to predict potential natural phenomena in the future vs describe the current
state of nature,
• Provide evidence in the near term vs the far term, and
• Provide needed evidence with the least effort and resources, including financial resources and
laboratory animals (efficiency).
9.2.10 Evaluating the Globalization Potential of GoF Benefits
Whether risks and benefits are equally distributed across populations is also an important consideration in
any risk-benefit comparison. For GoF research involving PPPs, the risks– that biosafety or biosecurity
incidents associated with the conduct of GoF research involving PPPs may spark a pandemic–are global.
In contrast, whether GoF benefits are globally distributed is likely to vary by the type of benefit
considered. The extent to which these benefits can be globalized influences whether risks and benefits are
equally distributed for a particular type of GoF study. To inform NSABB’s deliberations on this issue, the
benefit assessment qualitatively assessed the globalization potential of the latter set of GoF benefits,
through analysis of historical case studies examining the globalization of similar benefits and through
review of relevant USG policies (e.g. policies related to MCM sharing, etc.).
The globalization potential of select GoF benefits, namely those that are relevant worldwide but may be
primarily realized in the US and other developed countries, were evaluated. To support this task, USG
policies, programs, and international agreements relevant to globalization of GoF benefits were analyzed,
including USG policies and international agreements regarding MCM sharing during global outbreaks,
and other relevant pandemic preparedness support for WHO. Also, historical examples of USG
involvement in the globalization of GoF benefits were analyzed, considering the context of the historical
example and its relevance to a future outbreak of influenza, SARS or MERS. Taken together, these
analyses will enable qualitative assessment of the degree to which the USG promotes globalization of
various GoF benefits, as well as the timescale over which those benefits are expected to internationalize.

9.2.11 Quantitative Analysis of GoF Benefits
Although the ability to provide quantitative metrics for benefits would facilitate comparison of the
benefits of GoF versus alt-GoF research as well as of the risks and benefits associated with particular
types of GoF studies, given the differences in the availability and quality of data related to the realization
of the benefits, a quantitative analysis of all benefits cannot be performed. In particular, benefits related to
some aspects of MCM development, surveillance, public health policy, and scientific knowledge are
associated with multiple sources of uncertainty in how, when, and where the benefits will ultimately
improve the health of human populations, which precludes a meaningful quantitative analysis of the
magnitude of those effects. However, it is hoped that the rigorous examination of the pathways through
which those benefits lead to reductions in the burden of infectious diseases on human populations provide
a qualitative sense of the potential scale of each benefit, in light of current barriers to the realization of
that benefit.
Benefits related to the production of influenza vaccines are amenable to quantitative analysis, which
leverages models developed for the biosafety RA (specifically the nested SEIR models of global
outbreaks) to parametrically explore how changes in the control of outbreaks of PPP can mitigate
morbidity or mortality. Critically, many factors prevent the absolute assignment of a particular GoF
outcome to a quantitative benefit. For this reason, the quantitative approach herein shows how changing a
public health or medical capability that can be targeted by GoF research (such as the timeliness of the
availability of a vaccine during a pandemic) could affect the consequences of a global outbreak. These
data are accompanied by a commentary on the barriers for GoF achieving a desirable change to public
health and medical capabilities or preventing a deterioration of public health/medical capabilities so that
stakeholders can understand the probability of achieving the quantitative benefits modeled. This
quantitative component of the evaluation was accomplished using the HHS-BARDA Interactive Influenza
Model (as described in the biosafety RA described above) to parametrically analyze the effect of:
1. The timeliness of availability of a vaccine after an seasonal or pandemic influenza outbreak, and
2. The amount of vaccine available when it becomes available.
For each of these parameters, the value of the parameter was allowed to vary from arbitrarily large
numbers to arbitrarily small numbers during simulations of outbreaks of seasonal influenza and pandemic
influenza (similar to the pandemic strain of 1918 or 2009). This enabled determination of the value at
which each of these parameters begin to affect the consequence of global influenza outbreaks. The change
in parameter value needed to significantly change the consequence of a global outbreak was compared
with the plausible benefit to the vaccine afforded by GoF and alt-GoF studies, in order to determine if
either is likely to have the significant effect on the consequences of an outbreak. Moreover, for GoF
studies that are necessary to maintain the status quo for influenza vaccines, this analysis determined how
much worse an outbreak would be if those studies were not allowed to continue.
9.3 Coronaviruses: Benefits of GoF research
9.3.1 Summary
This section describes the benefits of GoF research involving coronaviruses (CoVs), which includes (1)
approaches that enhance virus production, (2) alter host range, (3) enhance virulence in appropriate
animal models, and (4) lead to evasion of therapeutics. Such GoF studies were found to generate scientific
knowledge, have direct applications to the development of vaccines and therapeutics, and may also have
economic benefits (not considered). Alt-GoF approaches that may generate similar benefits were also

identified and analyzed. At present, GoF studies involving CoVs have unique and direct benefits,
particularly to the development of vaccines and therapeutics.
9.3.1.1 GoF approaches that enhance virus production
9.3.1.1.1 Benefits to Scientific Knowledge
• GoF approaches that enhance virus production have potential to enable the development of in
vitro model systems for the study of any animal CoV in a variety of cell types, including
immortalized and primary cell lines. However, the fact that few animal CoVs identified to date
can be grown in existing cell culture systems limits the success of this approach.
9.3.1.2 GoF approaches that alter host range
• GoF approaches:
o Are uniquely capable of identifying novel viral genetic traits and factors that contribute to
cross-species adaptation, in any CoV strain,
o Are uniquely capable of demonstrating that a particular mutation(s) is necessary and

sufficient to alter the host range of a coronavirus,
o Enable the development of in vitro model systems for the study of any animal CoV in a
variety of cell types, including immortalized and primary cell lines, and
o Uniquely enable the development of animal models that recapitulate human disease
pathogenesis, which can be used to study many facets of disease pathogenesis, including the
role of viral and host immune factors in host pathology and the role of tissue tropism in
pathology.
• Alternative approaches:
o Comparative sequence analysis is uniquely capable of identifying genetic traits that are
associated with human adaptation, but this approach is limited to the study of CoVs that have
already caused human infections and is significantly constrained by the quality and
availability of genetic surveillance data for CoVs. In addition, the causality of mutations must
be confirmed through a GoF experiment.
o In vitro approaches, including characterization of the capacity of wild type viruses to infect
cells derived from various host species, the use of other viruses pseudotyped with CoV Spike
proteins, and binding assays using recombinant proteins, are limited to studying the role of
the Spike protein in cross-species adaptation. In addition, results using pseudotyped viruses
or recombinant proteins may not be recapitulated in the context of the wild type virus.
o Use of naturally permissive cell lines to study bat CoVs is limited to the few bat CoVs that
can productively infect and replicate within existing cell culture lines.
o Use of cell lines ectopically expressing permissive receptor proteins to study bat CoVs is
limited to cell lines that can be readily transfected, and modifications to cell lines may alter
the biology of infection.

o Naturally susceptible hosts of SARS and MERS cannot be used to study disease pathology
because they are asymptomatic or display different symptoms from humans.
o Transgenic animals that are expressing human receptor proteins do not recapitulate human
disease pathogenesis, thus results using transgenic animals may not translate to humans.
o Though human autopsy data provides direct information about human pathology, limited
autopsy data are available and mortalities are not representative of all cases, limiting the
generalizability of results.
o Alternative coronaviruses such as mouse hepatitis virus (MHV) can be used to gain insight
into basic aspects of CoV biology but are sufficiently distinct from human CoVs that they are
not suitable for the study of pathogenesis.
9.3.1.2.2 Benefits to Vaccine Development
• GoF approaches:
pathogenesis, which support testing of the safety and efficacy of candidate vaccines in a
robust system that can be used to demonstrate that vaccines reduce disease-associated
pathology and can reveal whether vaccines have adverse side effects, and
o Are uniquely capable of providing reliable information about the broad-spectrum potential of
CoV vaccines, through the use of chimeric bat-SARS CoVs as vaccine challenge viruses.
• Alt-GoF approaches:
o Other animal models (naturally susceptible hosts and transgenic animals) do not recapitulate
human disease pathogenesis, and thus are weak systems for demonstrating the efficacy of
vaccine candidates and cannot reveal adverse side effects.
o Few wild type bat CoVs can be cultured in existing cell lines, and bat CoVs do not naturally
infect mice, thus wild type bat CoVs have limited utility for the development of broad-
spectrum vaccines.
o Vaccine efficacy results using viruses pseudotyped with CoV Spike proteins must be
confirmed in wild type (or chimeric CoV strains) due to significant differences in the surface
presentation of Spike proteins.
9.3.1.2.3 Benefits to Therapeutic Development
• GoF approaches:
pathogenesis, which support testing of the safety and efficacy of candidate therapeutics in a
robust system that can be used to demonstrate that therapeutics reduce disease-associated
pathology,
o Are uniquely capable of providing reliable information about the broad-spectrum potential of
CoV therapeutics, through the use of chimeric bat-SARS CoVs as vaccine challenge viruses.

o Other animal models (naturally susceptible hosts and transgenic animals) do not recapitulate
human disease pathogenesis, and thus are weak systems for demonstrating the efficacy of
vaccine candidates and do not satisfy the FDA Animal Efficacy Rule.
9.3.1.3 GoF approaches that enhance fitness or virulence in cell culture or animal model systems
It should be noted that serial passaging of viruses in mice both alters the host range of the virus and
enhances its virulence in mice. The value of GoF benefits derived from the use of mouse-adapted viruses,
relative to alternative approaches, was summarized in Sec. 9.3.1.2 (GoF approaches that alter host range)
and will not be repeated in this section.
• GoF approaches:
o Represent the most efficient and effective strategy for identifying novel genetic traits and
viral factors that contribute to virulence, in any CoV strain, and
o Are uniquely capable of demonstrating that a particular mutation(s) is necessary and

sufficient to enhance the fitness/virulence of a coronavirus.
o Comparative sequence analysis is uniquely capable of identifying genetic traits that are
associated with enhanced virulence in humans but is limited to the study of SARS and MERS
and is significantly constrained by the quality and availability of genetic surveillance data for
CoVs. In addition, any hypotheses must be experimentally confirmed.
o Loss of function approaches (i.e. screening gene knockout viruses in vitro) are limited to the
discovery of viral factors involved in replication and may uncover factors that indirectly
contribute to virulence. Though targeted mutagenesis can be used to confirm that a genetic
trait is necessary for virulence, this LoF approach provides limited information about how
proteins cooperate to enhance virulence, which is a complex, multi-genic trait.
9.3.1.3.2 Benefits to Vaccine Development
• GoF approaches:
o Are uniquely capable of determining whether live attenuated vaccine viruses (LAVs) recover
virulence upon growth in cells or animals, a critical aspect of safety testing for this type of
vaccine, and
o Represent the most efficient and effective strategy for identifying novel virulence factors,
which can be deliberately attenuated to generate LAVs, a promising type of CoV vaccine
platform.
o Alternative experimental approaches for identifying virulence determinants are less efficient
than GoF approaches and are primarily limited to the study of known virulence factors,
limiting their utility for informing LAV development.
o Other types of vaccine platforms that do not rely on GoF approaches have strengths and
limitations relative to LAVs, which may rely on GoF for their development.

• GoF approaches:
o Represent the most efficient and effective strategy for identifying novel virulence factors,
which are potential therapeutic targets.
o Alternative experimental approaches for identifying virulence determinants are less
efficient that GoF approaches and are primarily limited to the study of known virulence
factors, limiting their utility for discovering potential new therapeutic targets.
o High-throughput screening of small molecule compounds for their ability to reduce viral
replication in vitro has generated promising therapeutic candidates, but such screens are
limited to the discovery of drugs that inhibit viral replication, only one aspect of
virulence.
o High-throughput screening of monoclonal antibodies (mAbs) for their ability to bind

CoV Spike proteins has generated promising therapeutic candidates, but mAb-based
therapeutics have several drawbacks, including the fact that CoV Spike proteins can
readily acquire escape mutations.
9.3.1.4 GoF approaches that lead to evasion of therapeutics in development
• GoF approaches:
o Are uniquely capable of determining the genetic threshold for resistance of a candidate
therapeutic prior to field deployment of the therapeutic, which is a recommended
component of an Investigational New Drug application to the FDA,
o Are uniquely capable of identifying the viral target of a novel therapeutic with an
unknown mechanism of action,
o Provide insight into the mechanism of activity of a therapeutic through the identification
of mutations that are necessary and sufficient to confer resistance to the therapeutic,
which is a recommended component of an Investigational New Drug application to the
FDA, and
o Are uniquely capable of determining the therapeutic dose that is least likely to lead to the
acquisition of antiviral resistance as well as determining whether combination therapies
better prevent the emergence of resistant viruses than individual therapies, which informs
the development of therapeutic strategies that will be effective for a longer time in the
field.
o X-ray crystallography and photoaffinity crosslinking are limited to the study of
therapeutics with known viral targets, and inferring mechanistic information based on
static data about drug-viral interactions may be difficult.

o RNAi screens to identify host factors that are required for the antiviral activity of a
therapeutic provide indirect information about the mechanisms of therapeutics that target
viral proteins.
GoF approaches that benefit the development of vaccines and therapeutics may lead to downstream
economic benefits, which were not analyzed in this report. GoF approaches involving coronaviruses do
not benefit surveillance, informing policy decisions, or the development of diagnostics.
9.3.2 Overview of the GoF research landscape involving coronaviruses
This assessment describes the benefits of GoF experiments involving SARS-CoV, MERS-CoV, and
SARS/MERS-like bat CoVs. From a review of the coronavirus literature, experimental approaches were
identified that are reasonably anticipated to lead to the following phenotypic changes:

• Altered host range (typically accompanied by enhanced virulence in the new host),
• Enhanced fitness or virulence in cell culture or laboratory animal model systems respectively, and
natural or induced immunity. Finally, no coronavirus research that is reasonably anticipated to lead to
evasion of diagnostics or of vaccines in development was identified. (Additionally, there are currently no
FDA-approved vaccines or therapeutics for coronaviruses.)
Of note, the four human coronaviruses that cause mild to moderate respiratory illnesses such as the
common cold or croup (coronaviruses HKU1, OC43, 229E, and NL63) were not evaluated because these
are not considered in the NSABB GoF Framework. Throughout this report, the use of the term
“coronaviruses” or “CoVs” refers specifically to SARS-CoV, MERS-CoV, and SARS/MERS-like bat
CoVs such as HKU4 and HKU5.
The following chapter summarizes the results of the assessment of the benefits of GoF research involving
coronaviruses. A more detailed analysis to further support the findings described in Chapter 9.3 is
presented in Appendix IV Sec. 15.1. As the relative ability of a given GoF (or alt-GoF) approach to
address a particular scientific knowledge or public health gap often hinges on nuanced differences
between the benefits and limitations of different approaches, readers who seek an in-depth understanding
of the benefits of GoF research are directed to chapter 15.
In the following section, a brief overview of the experimental approaches within each GoF phenotypic
category is provided and the scientific outcomes and/or products of each approach are described.
9.3.2.1 Experimental approaches that lead to enhanced pathogen production
Serial passaging of CoV in cell culture leads to the generation of higher-yield viruses. This approach is
used to enhance the growth of viruses with naturally poor growth properties, in order to develop an in
vitro model system for experimental use.

9.3.2.2 Experimental approaches that alter host tropism in mammals
A second approach that leads to altered host range involves serial passaging of CoVs in mice, which leads
to the generation of viruses that have adapted to more efficiently infect and cause disease in mice. The
purpose of this experiment is two-fold:
A final approach involves targeted mutagenesis to introduce mutations that are associated with altered
host tropism, which is performed to demonstrate that the mutation(s) are necessary and sufficient to alter
host tropism. This information provides a foundation for follow-up studies investigating the phenotypic
traits underlying virus adaptation to new hosts.
9.3.2.3 Experimental approaches that enhance fitness or virulence in cell culture or laboratory animal
model systems

and sufficient to enhance virulence. This information provides a foundation for follow-up studies to
elucidate the mechanistic basis of virulence.
A third approach involves serial passaging of attenuated viruses that are candidate live attenuated
vaccines, in order to determine whether the viruses acquire mutations that enhance fitness/virulence.
Because LAVs with an ability to recover fitness during growth in vivo could cause adverse outcomes in
people, a negative result is an important indicator of safety for any live attenuated vaccine in
development.
9.3.2.4 Experimental approaches that lead to evasion of therapeutics in development
that are resistant to inhibition/neutralization by that therapeutic. The purpose of the experiment is to
understand whether and how readily resistance will arise in response to selective pressure from the
therapeutic and to identify mutations that are associated with resistance to the therapeutic, which provides
a foundation for follow-up studies investigating the mechanisms underlying antiviral activity and antiviral
resistance. This information benefits the development of these therapeutics. Specifically, emergence-of-
resistance data speaks to the potential field efficacy of the therapeutic, and information on both antiviral
mechanism and emergence of resistance are important components of an investigational new drug
application to the FDA.
9.3.3 Identification of potential benefits and limitations of GoF research involving CoVs
In this section, the potential benefits of GoF research involving CoVs in each benefit category listed in the
NSABB Framework are evaluated.
9.3.3.1 Benefits and Limitations of GoF to Scientific Knowledge
9.3.3.1.1 Scientific Knowledge Benefit #1: Gain insight into the mechanisms underlying adaptation of
animal CoVs to humans
SARS and MERS unexpectedly emerged from their animal reservoirs to infect humans in 2002 and 2012,
respectively. Surveillance of bats and other CoV reservoir species indicates that there is a large diversity
of animal CoVs circulating in nature, including many species that are genetically related to SARS and

MERS and thus may have the potential to spill over into human populations in the future.478,479,480,481
Although multiple coronaviruses have been shown to exhibit a flexible capacity for cross-species
transmission,482,483 the mechanisms underlying CoV adaptation to new host species are poorly understood.
Several GoF approaches have potential to address this scientific knowledge gap. Serial passaging of CoVs
in cells derived from a non-natural host organism or in a non-natural laboratory animal host selects for
viruses that more efficiently infect cells/animals, thereby enabling the identification of mutations that are
sufficient for adaptation to a new host species. Identifying where mutations arise during adaptation to new
hosts points to viral factors that may play a role in adaptation, and studying the phenotypic consequences
of the mutations provides insight into the mechanistic basis of cross-species adaptation. One key benefit
of this approach is that it can lead to the discovery of novel genetic traits and virus proteins that are
involved in the process of adapting to new hosts without the need for prior knowledge of viral adaptation
factors. Moreover, this approach can be used to explore the adaptation of any virus to a new host species,
provided that the virus can be grown in an appropriate model system. The main limitation of this
approach is that laboratory results in cell culture or animal model systems may not translate to viral
adaptation to humans in nature. Additionally, results gleaned from the one or two strains under study may
not be conserved in other CoV species.
Another GoF method for studying cross-species adaptation involves “Spike swapping” – that is, targeted
genetic modification to replace all or part of the CoV Spike protein, a surface protein that mediates virus
entry into cells and is a critical determinant of host restriction, with the Spike protein from another CoV
species. These experiments are considered gain-of-function because they are expected to alter host
tropism in mammalian species. The purpose of these experiments is two-fold. First, testing the ability of
chimeric CoVs to infect various types of cells and animals reveals the breadth of host tropism conferred
by a given Spike protein, and comparing the sequences of parental and donated Spike proteins with
different host tropism can uncover amino acid residues that mediate host restriction. Second, defining the
host tropism of animal CoVs and the number of amino acid changes that are needed to confer the ability
to infect human cells provides insight into whether the ability to adapt to new species is a conserved
feature of CoVs, as well as which animal CoVs are poised to spill over into human populations. Third,
because most bat CoVs cannot be cultured in standard cell culture systems, “Spike swapping” enables the
chimeric bat-SARS virus to infect and replicate within human cells, thereby enabling further study of the
behavior of the bat CoV. The main drawback of this approach is that it is limited to studying the role of
the Spike-receptor interaction in host tropism. Another drawback is that chimeric “SARS plus animal
CoV Spike” viruses may behave differently from wild type animal CoVs.
A third GoF approach involves serial passaging of bat CoVs in cell culture, which selects for viruses that
are better able to bind, infect, and replicate within human cells (i.e. enhanced pathogen production). For
those bat CoVs that can infect cells but grow poorly in cell culture, this enables the development of
higher-yield viruses that can be used as tools for the study of bat CoV behavior. Understanding the
479 Yang Y et al (2015) Two Mutations Were Critical for Bat-to-Human Transmission of Middle East Respiratory Syndrome
Coronavirus. Ibid. 89: 9119-9123
480 Pfefferle S et al (2009) Distant relatives of severe acute respiratory syndrome coronavirus and close relatives of human
coronavirus 229E in bats, Ghana. Emerging infectious diseases 15: 1377-1384
481 Ge XY et al (2013) Isolation and characterization of a bat SARS-like coronavirus that uses the ACE2 receptor. Nature 503:
535-538
482 Baric RS et al (1999) Persistent infection promotes cross-species transmissibility of mouse hepatitis virus. Journal of
virology 73: 638-649
483 Chen W et al (2005) SARS-associated coronavirus transmitted from human to pig. Emerging infectious diseases 11: 446-
448

characteristics of bat CoVs relative to human epidemic CoVs may provide insight into the adaptive
changes that facilitate efficient infection of humans.
Finally, targeted genetic modification of wild type viruses to introduce mutations that are associated with
adaptation to new hosts demonstrates that such markers are necessary and sufficient to broaden or alter
host tropism. This information provides a strong foundation for follow-up studies investigating the
mechanistic basis of the adaptation phenotype.
9.3.3.1.2 Scientific Knowledge Benefit #2: Gain insight into the mechanisms underlying the pathogenicity
of CoVs
Why SARS and MERS coronaviruses cause severe respiratory infections while other human
coronaviruses cause mild to moderate illness is unknown.484 Specifically, the viral genetic and phenotypic
traits underlying the enhanced pathogenicity of SARS and MERS relative to other human coronaviruses
are poorly understood, and only a few viral virulence factors have been identified and characterized (such
as the CoV Spike protein, which mediates viral entry into host cells).
Serial passaging of CoVs in cell culture or laboratory animals, which selects for enhanced fitness (in
vitro) or enhanced virulence (in vivo), is a GoF approach that can yield information that addresses this
scientific knowledge gap. This approach enables the identification of mutations associated with enhanced
fitness/virulence, which can lead to the discovery of new viral virulence factors and provides a foundation
for follow-up studies investigating the mechanistic basis of the enhanced fitness/virulence phenotype
observed in emergent viruses. A key benefit of this approach is the ability to generate and identify novel
mutations and viral proteins that contribute to fitness/virulence, without prior knowledge about viral
virulence factors. Moreover, this approach can be performed with any coronavirus that is capable of
infecting appropriate cell culture or animal model systems. The main drawbacks of serial passaging
experiments are that insights may not translate to human infections, and viral factors and phenotypes that
contribute to virulence in the CoV strain under study may not generalize to other CoV strains.
A second GoF approach for studying virulence involves targeted genetic modification of wild type viruses
to introduce mutations that are associated with enhanced fitness/virulence, which demonstrates that such
markers are necessary and sufficient to enhance fitness/virulence. This information provides a strong
foundation for follow-up studies investigating the mechanistic basis of the enhanced virulence phenotype.
9.3.3.1.3 Scientific Knowledge Benefit #3: Gain insight into disease pathogenesis, including host factors
that contribute to disease pathology
The host factors involved in SARS and MERS pathogenesis are poorly understood. That is, the
contribution of host immune responses to the exacerbated pathology observed during infection with
SARS-CoV and MERS-CoV relative to the “common cold” human CoVs is unknown.
Animal-adapted viruses, generated through serial passaging of CoVs in mice to enhance their capacity to
infect and cause disease in mice (i.e. altered host range and enhanced virulence) are essential tools for the
study of CoV pathogenesis. Infection of mice with animal-adapted viruses recapitulates disease pathology
observed during human infection, which is critical for studying the mechanisms underlying disease
pathology. Many different experimental methods can be used to study disease pathology using mouse
models, including characterizing the host immune response to CoV infection, knocking out or depleting
specific host immune factors to probe their role in pathogenesis, and analyzing the tissue tropism and
dissemination of CoVs over the course of infection. Of note, mouse-adapted viruses are also important for
the study of viral genetic and phenotypic traits that contribute to pathogenesis (scientific knowledge gap
484 (2015i) Interviews with coronavirus researchers.

#2). The main drawback of using mouse-adapted viruses is that adaptive changes may alter the biology of
the virus, such that findings are mis-representative of wild type virus behavior.
9.3.3.2 Benefits and Limitations of GoF to Surveillance
Currently, GoF approaches do not have the potential to benefit public health, agricultural animal, or
wildlife surveillance. Although CoV researchers stated that they could envision using information about
the molecular determinants of human adaptation and virulence to assess the risk posed by animal CoVs
circulating in nature, similar to the influenza field, this application is currently unfeasible for two reasons:
(1) CoV surveillance networks are extremely limited, with large gaps in coverage in humans and animals,
and (2) the state of knowledge about the molecular determinants of human adaptation and virulence is
poor.485
9.3.3.3 Benefits and Limitations of GoF to Vaccine Development
Currently, there are no FDA-approved vaccines for CoVs, which represents a critical gap in public health
preparedness for CoV outbreaks. Several GoF approaches have the potential to benefit the development
of new CoV vaccines.
9.3.3.3.1 Vaccine development benefit #1: Developing vaccine candidates
GoF approaches have the potential to benefit two aspects of the development of live attenuated vaccine
(LAV) platforms, which is a type of vaccine that is being actively researched for its potential as a CoV
vaccine platform. First, GoF approaches can inform the development of candidate LAV strains, which
exhibit attenuated virulence relative to parental strains. Specifically, one strategy for generating LAV
strains is through serial passaging in a non-human host (either an animal or cells derived from an animal),
as adapting a virus to a new host typically attenuates the virus in humans (i.e. alters rather than enhances
host tropism). Because this approach alters host tropism, it is considered to be a GoF approach under the
NSABB Framework. Although serial passaging has been used historically for developing polio, smallpox
and other viral vaccines, the approach has not been utilized for the purpose of developing CoV vaccine
strains.486 Alternatively, live attenuated vaccines can be generated through targeted mutagenesis to
attenuate or knock out the function of known virulence factors. As described above (Sec. 9.3.3.1.2), GoF
studies that enhance virulence represent the most efficient and effective strategy for identifying novel
CoV virulence factors, which may be good targets for attenuation to develop an LAV. However, follow-
up studies are needed to determine how to attenuate that factor or to render it non-functional.
LAVs are an appealing type of vaccines for CoVs for several reasons, and multiple LAV candidates for
SARS have been shown to protect against lethal virus challenge in mice, demonstrating the promise of
this type of vaccine for CoVs.487,488 However, a major concern is their potential to regain virulence in
people, which necessitates stringent safety testing of all LAV candidates.
485 For example, out of more than 1700 bat species, only ten have been surveilled for evidence of CoV infection (and those ten
on an ad hoc rather than a systematic basis).
486 Ulmer JB et al (2006) Vaccine manufacturing: challenges and solutions. Nature biotechnology 24: 1377-1383
487 Graham RL et al (2012) A live, impaired-fidelity coronavirus vaccine protects in an aged, immunocompromised mouse
model of lethal disease. Nature medicine 18: 1820-1826
488 Fett C et al (2013) Complete protection against severe acute respiratory syndrome coronavirus-mediated lethal respiratory
disease in aged mice by immunization with a mouse-adapted virus lacking E protein. Journal of virology 87: 6551-6559

9.3.3.3.2 Vaccine development benefit #2: Determining the potential for LAVs to recover virulence.
Once a candidate LAV strain has been generated, the strain is typically serially passaged in vitro or in
vivo to determine whether the virus recovers fitness/virulence (i.e. enhanced fitness/virulence). Because
a tendency to revert or acquire compensatory mutations that enhance fitness/virulence could seriously
compromise the safety of a live attenuated vaccine, demonstrating the genetic stability of a candidate
LAV is a critical aspect of its development.
9.3.3.4 Benefits and Limitations of GoF to Therapeutic Development
Currently, there are no FDA-approved therapeutics for CoVs, which represents a critical gap in public
health preparedness for CoV outbreaks. Several GoF approaches have the potential to benefit the
development of new CoV therapeutics.
9.3.3.4.1 Therapeutic development benefit #1: Developing candidate therapeutics
CoV researchers cited the lack of knowledge of good viral targets for therapeutics as a critical limitation
for the development of CoV therapeutics.489 GoF approaches currently represent the most efficient and
effective way to identify novel virulence factors and gain insight into their mechanism of activity, a
foundation for the development of antivirals (see Sec. 9.3.3.1.2). However, whether inhibiting or
attenuating the virulence factor is sufficient to reduce viral replication and infection-associated pathology
must be determined through alternative approaches.
9.3.3.4.2 Therapeutic development benefit #2: Generating nonclinical data to support an Investigational
New Drug application to the FDA
The first step in the licensure process for new drugs involves submission of an Investigational New Drug
(IND) application to the FDA’s Center for Drug Evaluation and Research (CDER). CDER recommends
that several types of nonclinical studies are conducted before starting Phase I clinical studies, including
determination of the drug’s mechanism of action, in vitro selection of resistant viruses to the
investigational product, and the genotypic and phenotypic characterization of resistant viruses.490 GoF
approaches that lead to evasion of therapeutics generate information that fulfills both of those
recommendations, thereby supporting the licensure of new therapeutics.
First, serial passaging of viruses in the presence of a therapeutic to select for resistant viruses, followed by
sequencing of the emergent resistant strains to identify genetic changes that arose, can provide insight into
the mechanism of action of the therapeutic. Understanding which viral protein or proteins mutate in order
for the virus to escape inhibition suggests those proteins are targeted by the therapeutic, and the site and
phenotypic consequences of the mutations may provide insight into the mechanism of antiviral activity.
Together, this information provides a foundation for follow-up structural, biochemical, and cell biological
assays investigating the mechanism of antiviral activity. Second, this approach directly fulfills FDA’s
recommendation for in vitro selection of resistant viruses, which is performed to determine the genetic
threshold for the development of resistance (i.e. the number of mutations that are needed for a virus to
acquire resistance).

490 Food and Drug Administration. Guidance for Industry: Antiviral Product Development - Conducting and Submitting
Virology Studies to the Agency.
http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM070953.pdf. Last
Update June 2006. Accessed 14 October 2015.

9.3.3.4.3 Therapeutic development benefit #3: Determining the therapeutic dosage and/or combination
therapies that are least likely to lead to the emergence of resistance
GoF studies that lead to evasion of therapeutics can also inform the therapeutic dosage and the use of
combination therapies, both of which influence whether and how readily antiviral resistance arises.
Specifically, serial passaging of virus in animals dosed with varying amounts of the therapeutic provides
insight into the dosage that is least likely to lead to the emergence of resistant viruses, and serial
passaging of virus in cells or in animals in the presence of multiple mAbs (or other types of therapeutics)
can be used to determine how readily resistance arises in response to combination versus single therapies.
This information may lead to the development of therapeutic strategies that will be effective for a longer
period of time in the field.
9.3.3.5 Benefits and Limitations of GoF to Both Vaccine and Therapeutic Development
9.3.3.5.1 Vaccine/Therapeutic Development Benefit #1: Testing the safety and efficacy of MCM
candidates
The use of animal-adapted viruses, generated using GoF approaches that alter host range and enhance
virulence, facilitate MCM development by enabling the testing of MCM candidates in an animal model
that mimics the pathology of human disease. Animal-adapted strains represent a robust system for
demonstrating that a candidate MCM is capable of preventing or reducing disease-associated pathology.
In addition, the use of models that share features of human disease can reveal adverse side effects of the
vaccine and thus is an important aspect of safety testing prior to the initiation of human clinical trials.
9.3.3.5.2 Vaccine/Therapeutic Development Benefit #2: Developing broad-spectrum vaccines
Finally, GoF approaches that alter host range inform the development of broad-spectrum vaccines that
may be capable of protecting against the next emerging CoV. Specifically, chimeric bat-SARS viruses
can be used as challenge viruses to explore the broad-spectrum potential of candidate MCMs, in order to
test whether MCMs designed to target SARS/MERS proteins are also capable of targeting cognate
proteins in bat CoVs as well as whether vaccines can target SARS/MERS proteins in a different virus
context (representative of the next emerging CoV capable of infecting humans). These experiments can
provide insight into whether MCMs targeting any CoV protein or process are capable of conferring
broad-spectrum protection against bat CoVs with zoonotic potential, in addition to SARS and MERS.
9.3.3.6 Benefits and Limitations of GoF to Diagnostic Development
As diagnostic targets for CoVs are well-established, potential benefits of GoF approaches to the
development of diagnostics were not identified.491, 492,493, 494
491 The FDA-approved diagnostic test for MERS-CoV targets two regions in the CoV genome: a region upstream of the E gene
(upE) and the reading frame 1a (orf1a). SARS can be detected through RT-PCR with sequences in the polymerase 1 B
region (pol 1B) and an adjacent downstream region of the genome as the targets. Other diagnostic tests target sequences in
the nucleocapsid (N) gene.
492 Stephen M. Ostroff Acting Commissioner of Food and Drugs. Letter of Authorization RealStar® MERS-CoV RT-PCR Kit
U.S. . http://www.fda.gov/downloads/MedicalDevices/Safety/EmergencySituations/UCM455348.pdf. Last Update July 17,
2015. Accessed December 2015.
493 Richardson SE et al (2004) The laboratory diagnosis of severe acute respiratory syndrome: emerging laboratory tests for an
emerging pathogen. The Clinical biochemist Reviews / Australian Association of Clinical Biochemists 25: 133-141
494 Mahony JB et al (2004) Performance and Cost evaluation of one commercial and six in-house conventional and real-time
reverse transcription-pcr assays for detection of severe acute respiratory syndrome coronavirus. J Clin Microbiol 42: 1471-
1476

9.3.3.7 Benefits and Limitations of GoF to Decision-Making in Public Health Policy
Because the U.S. government is not actively engaged in public health preparedness activities for CoV
outbreaks and because there are no FDA-approved vaccines or therapeutics for CoVs, GoF approaches do
not have the potential to benefit decision-making in public health policy (e.g. informing countermeasure
stockpiling decisions, guiding decisions about strain selection for vaccine development, etc.)
9.3.3.8 Economic Benefits
GoF benefits to the development of vaccines and therapeutics could have downstream economic benefits.
Economic benefits were not explicitly evaluated in this report.
9.3.4 Identification of alt-GoF that provide similar potential benefits to the GoF being examined
In this section, an overview of alternative (alt-GoF) approaches that yield the same or similar benefits as
the GoF approaches described above is provided. Two types of alt-GoF approaches are reviewed: (1)
alternative experimental approaches that can provide the same or similar scientific information as GoF
experimental approaches, and (2) alternative scientific and technical innovations that can yield the same
public health benefits as GoF approaches but through different mechanisms, including the use of
alternative model systems that do not rely on GoF approaches. For each approach, the scientific outcomes
or products of the approach are first described, then how that information or products leads to similar
benefits as GoF approaches.
9.3.4.1 Benefits and Limitations of Alt-GoF Approaches to Scientific Knowledge
Several alternative experimental approaches can be used to discover genetic traits associated with cross-
species adaptation of CoVs.
Comparing the sequences of CoVs with different species tropism, including comparison of animal CoVs
versus SARS/MERS and comparison of animal strains from different geographic regions where spillover
into human populations has and has not occurred (or has occurred with different frequencies), can
elucidate genetic traits that are associated with adaptation to different hosts. Second, comparative
sequence analysis of human CoVs from different time points during an outbreak reveals how zoonotic
CoVs adapt to humans following an initial spillover event. Relative to the laboratory methods described
above, this approach has potential to identify traits that are relevant for adaptation to humans under
natural selective pressures. Importantly, follow-up studies are needed to confirm that the identified
genetic traits are responsible for altered host tropism.
Both types of comparative sequence approaches suffer from several significant limitations. First, the
success of comparative sequence analysis is significantly constrained by the quality and availability of
existing genetic surveillance data. A second limitation is that, due to the large size of the CoV genome
(27-32 kb) and the genetic diversity of coronaviruses in nature, there are a very large number of genetic
differences between any two CoV strains, only a subset of which are likely to be important for cross-
species adaptation.495 Because of that “noise,” sequence comparisons are realistically limited to known
regions of interest, precluding discovery of novel factors that are involved in host adaptation. Due to the

fact that only a few proteins have been shown to be involved in cross-species adaptation and the function
of most CoV proteins is unknown, this limited focus represents a critical shortcoming of the comparative
sequence analysis approach. Although this limitation could be partially addressed by comparing
sequences of paired animal and human isolates, few such paired sequences are available. Third, this
approach is reactive, limited to the study of mechanisms underlying adaptation of CoVs that have already
evolved to broaden or alter their host tropism (e.g. SARS and MERS). The mechanisms driving
adaptation of other CoVs to new hosts may be different.
Several alternative approaches seek to define the breadth of host tropism conferred by a given Spike
protein. The first approach involves testing whether MERS- or SARS-CoVs can infect cells derived from
various non-human host species such as bats or cells that do not naturally express CoV receptor proteins
but have been engineered to ectopically express receptor proteins from various species. This approach
cannot be used for most animal CoVs, which cannot be grown efficiently in cell culture to produce
infectious material for laboratory assays. Alternatively, two virus-free approaches can provide
information about compatible Spike-host interactions: (1) in vitro binding assays using recombinant Spike
proteins and host receptor proteins from different species, and (2) cell culture-based binding and virus
entry assays using non-CoVs (e.g. murine leukemia virus) that are pseudotyped with CoV Spike proteins.
(Pseudotyping is the process of expressing the envelope protein or surface glycoprotein from one virus on
the surface of a different virus, e.g. replacement of the vesicular stomatitis virus glycoprotein (VSV G)
with the CoV Spike, enabling expression of the CoV Spike on the surface of VSV.) These in vitro
systems can also be used to confirm that amino acid substitutions in the Spike protein are necessary and
sufficient to alter host receptor binding and cell entry capabilities. The major limitation associated with
these virus-free approaches is that results may not be recapitulated in the context of the wild type virus, as
the virus context influences presentation of surface epitopes. Additionally, results from either virus-free
approach may not be conserved in a different strain context, and traits that promote binding of
pseudotyped viruses to a particular cell type may not be critical for adaptation to human hosts. Finally,
these approaches are currently used to investigate the role of the Spike-receptor interaction in host
restriction only.
Finally, structural modeling of Spike-receptor interactions, based on crystal structures of Spike-receptor

complexes, can also be used to identify amino acid residues in the Spike protein that may be important
determinants of host restriction. Though useful for generating hypotheses about mutations that may alter
host tropism, all predictions must be experimentally confirmed.
In addition to the use of serially-passaged bat CoVs or chimeric CoVs, several alternative model systems
can be used to study the biology of bat CoVs, which may provide insight into the adaptive changes that
are needed for CoVs to efficiently infect humans. First, some bat CoVs are naturally capable of
replicating within bat cell lines or other standard cell culture systems. However, bat cell lines are much
less experimentally tractable than human cell lines, as fewer reagents are available and the cells are more
difficult to transfect than human cells, further lessening the utility of this approach. 496,497 Second, host
cells that are not naturally permissive to infection with animal CoVs can be sensitized to infection through
ectopic expression of the receptor protein from the natural host species (or another permissive host
species). This approach has been utilized for a limited number of CoVs, and whether it will permit
replication of a broad range of emerging CoVs is unknown. Furthermore, this strategy cannot be used for
primary cell lines, which are not readily transfectable, and overexpression of the receptor may alter the
process of infection, leading to artefactual results.
497 Huynh J et al (2012) Evidence supporting a zoonotic origin of human coronavirus strain NL63. Ibid. 86: 12816-12825

of CoVs
Several alternative approaches can also be used to study pathogenicity. Comparative sequencing of
SARS-CoV and MERS-CoV epidemic strains with varying levels of virulence can lead to the
identification of mutations associated with enhanced virulence. A strength of this approach relative to
serial passaging is that comparative sequence analysis uncovers genetic variation that is specially
associated with enhanced virulence in humans.498,499 However, this approach is limited to CoVs that have
already produced epidemics in humans, i.e. SARS-CoV and MERS-CoV. The success of this approach is
constrained by the quality and availability of surveillance data, in particular the quality of “metadata”
about clinical severity that is needed to “bin” sequences into low- and high-virulence categories for
comparison. While SARS-CoV strains from the early, middle, and late phases of the 2002 – 2003
epidemic have been found to exhibit varying levels of virulence (and have been used for comparative
sequence analysis studies), genetic surveillance data for MERS are limited. Finally, given the large size of
the CoV genome and genetic diversity among wild type CoV sequences, sequence comparisons are
practically limited to pre-determined regions of interest, which precludes identification of novel virulence
factors.
A second sequence-based approach involves analyzing the evolution of CoVs over time. Understanding
which regions of the genome mutate and which do not can provide insight into which regions are likely to
be critical for the virus life cycle, which may or may not contribute to pathogenicity. However, the utility
of this approach is also limited by the number of available CoV sequences.
Loss-of-function (LoF) studies, which involve knocking out or otherwise hampering the function of a
gene of interest (or its product) and screening for attenuated fitness (in vitro) or virulence (in vivo),
represent another alternative approach for the discovery of viral virulence factors and genetic traits
associated with virulence. Given the large size of the CoV genome, a random mutagenesis approach is
practically limited to the investigation of known virulence factors. A targeted gene knockout strategy can
be used to identify new viral genes that contribute to virulence, but a limited number of mutants can be
screened for attenuated virulence in vivo, due to the labor, expense, and ethical considerations associated
with the conduct of animal experiments. Thus, high-throughput screening of gene knockout viruses is
limited to screening for attenuated fitness in cell culture systems, which is only one aspect of virulence.
The major drawback of LoF screens is that losing the functionality of a virus protein, either through gene
knockout of mutagenesis, may indirectly attenuate virulence, so that gaining meaningful information
about virulence mechanisms may be difficult using this approach. Finally, it is noted that knocking out
the function of an unknown viral protein can lead to a loss or gain of virulence, depending on the function
of the protein.
LoF approaches can also be used to confirm that a particular trait is necessary for enhanced virulence.
However, because virulence is a complex, multi-genic trait, knocking out the function of one gene or
introducing a mutation into one gene may be sufficient to attenuate virulence but provides an incomplete
picture of the role of that particular protein. Additionally, mutations that are found to enhance virulence in
model systems may not translate to increased virulence during human infections.
498 de Jong MD et al (2005) Oseltamivir Resistance during Treatment of Influenza A (H5N1) Infection. New England Journal
of Medicine 353: 2667-2672
499 Chinese SMEC (2004) Molecular evolution of the SARS coronavirus during the course of the SARS epidemic in China.
Science 303: 1666-1669

In addition to using animal-adapted viruses generated through GoF approaches, several alternative model
systems can be used to study disease pathogenesis.
Naturally susceptible laboratory animals represent one alternative model system for studying disease
pathogenesis. However, laboratory animals that are naturally susceptible to infection with SARS-CoV and
MERS-CoV have been found to support viral replication but remain asymptomatic or develop symptoms
dissimilar to those in humans. Thus, these animal models are not suitable for pathogenesis studies.
Use of transgenic animals expressing the human virus receptor is another alternative to the use of adapted
viruses for hosts that are not permissive to infection or do not recapitulate human disease pathology. A
variety of approaches have been used to create transgenic mouse models for SARS-CoV and MERS-CoV
infection, and each technique results in a slightly different gene expression pattern and reproduces human
disease symptoms to a different degree. The ability to infect transgenic mice with wild type SARS-CoV
and MERS-CoV is a strength of this model system. However, given differences in pathogenesis, results
may not translate to human disease.
Finally, human autopsy data can be an alternative source of pathogenesis information. However, the
availability of these data are limited – autopsies are not often performed in Middle Eastern cultures, and
data has not yet been shared from the most recent outbreak in the Republic of Korea.500 Furthermore,
analysis of human autopsy data provides limited mechanistic insight because it is inherently correlative
and is devoid of time series information, obscuring the order in which pathogenic effects occurred.
Additionally, insights gleaned from the study of severe, end-stage disease may not be representative. The
fact that many SARS-CoV and MERS-CoV deaths occurred in patients with pre-existing conditions
further complicates the identification of pathology caused by viral infection versus comorbidities.
9.3.4.2 Benefits and Limitations of Alt-GoF Approaches to Vaccine Development
Live attenuated vaccines (LAVs) may be generated through targeted mutagenesis to knock out or
attenuate the function of known virulence factors. LoF approaches, namely screening of gene knockout
viruses or randomly mutagenized viruses for attenuated virulence, are relatively inefficient for the
discovery of novel virulence factors but can be used to confirm that inhibiting or attenuating the function
of a virulence factor is sufficient to attenuate virus replication.
In addition to LAVs, several other types of CoV vaccines are in development, which do not rely on GoF
approaches for their initial development. Alternative vaccine platforms of interest include inactivated
whole virus vaccines, recombinant vaccines, DNA vaccines, viral vector-based vaccines, and virus-like
particles (VLPs).501 Many of these vaccine types have shown promise, and each has strengths and
limitations relative to the use of live attenuated vaccines.
There are no alternative approaches for determining the potential for LAVs to recover virulence upon
growth in cells or animals prior to the clinical testing of vaccine candidates in people.
501 Zhang N et al (2014) Current advancements and potential strategies in the development of MERS-CoV vaccines. Expert Rev
Vaccines 13: 761-774

9.3.4.3 Benefits and Limitations of Alt-GoF Approaches to Therapeutic Development
Several alternative approaches can inform the development of candidate therapeutics against CoVs. First,
LoF approaches can lead to the identification of novel virulence factors, which may be good targets for
new therapeutics. LoF approaches are relatively inefficient for the discovery of novel virulence factors
but are critical for demonstrating that inhibition or attenuation of a virulence factor is sufficient to reduce
viral replication or infection-associated pathology.
An alternative approach to the targeted development of therapeutics involves high-throughput screening

of compounds for their ability to reduce viral replication in vitro.502,503,504,505,506 This is also an active area
of therapeutic research in the CoV field and has generated several promising candidates. One drawback of
this approach is that it is limited to the identification of compounds that reduce viral replication, which is
only one aspect of virulence. Targeting other aspects of virulence, such as viral interactions with the host
immune system, may prove to be a more effective therapeutic strategy.
A related alternative approach involves high-throughput screening of panels of monoclonal antibodies

(mAbs) to identify mAbs that bind to CoV Spike proteins, as mAbs targeting the Spike protein have been
shown to effectively prevent viruses from infecting cells and could prime the immune system to clear the
infection.507 One potential drawback of this therapeutic strategy is that CoVs can readily acquire
mutations in their Spike protein that enable escape from mAb neutralization; however, researchers are
actively pursuing the development of “cocktails” of mAbs that are more robust to the generation of escape
mutants.508,509 Additional drawbacks are that antibody-based therapeutics, which are uncommon for
infectious diseases, may only slow infections and must be injected because antibodies are not small
molecules.
Several alternative approaches can be used to investigate the mechanism of activity of a new therapeutic
candidate. First, high-throughput RNAi screens targeting host proteins can identify host proteins that are
required for the drug’s mechanism of action, by demonstrating that knockdown of a particular host
protein impedes the drug’s ability to inhibit viral replication. Though an informative strategy for the study
of therapeutics targeting host proteins, high-throughput RNAi screens provide minimal information about
potential viral targets of therapeutics. (It should be noted that this approach is typically performed to
502 de Wilde AH et al (2014) Screening of an FDA-approved compound library identifies four small-molecule inhibitors of
Middle East respiratory syndrome coronavirus replication in cell culture. Antimicrobial agents and chemotherapy 58: 4875-
4884
503 Dyall J et al ibid.Repurposing of clinically developed drugs for treatment of Middle East respiratory syndrome coronavirus
infection. 4885-4893
504 Ratia K et al (2008) A noncovalent class of papain-like protease/deubiquitinase inhibitors blocks SARS virus replication.
Proceedings of the National Academy of Sciences of the United States of America 105: 16119-16124
505 Wu CY et al (2004) Small molecules targeting severe acute respiratory syndrome human coronavirus. Ibid. 101: 10012-
10017
506 Severson WE et al (2007) Development and validation of a high-throughput screen for inhibitors of SARS CoV and its
application in screening of a 100,000-compound library. Journal of biomolecular screening 12: 33-40
507 Sui J et al (2008) Broadening of neutralization activity to directly block a dominant antibody-driven SARS-coronavirus
evolution pathway. PLoS pathogens 4: e1000197
508 Rockx B et al (2010) Escape from human monoclonal antibody neutralization affects in vitro and in vivo fitness of severe
acute respiratory syndrome coronavirus. The Journal of infectious diseases 201: 946-955
509 Sui J et al (2014) Effects of human anti-spike protein receptor binding domain antibodies on severe acute respiratory
syndrome coronavirus neutralization escape and fitness. Journal of virology 88: 13769-13780

identify the potential targets of drugs identified through high-throughput screens, as the candidate drugs
may attenuate viral replication by directly targeting viral proteins or by indirectly targeting host proteins.)
If the therapeutic target of a drug is known, analyzing the crystal structure of the viral target in complex
with the antiviral compound (or mAb) can provide insight into the compound’s mechanism of
activity.510,511 This approach is particularly useful for therapeutics that directly bind to and inhibit the
activity of a viral protein. Though X-ray crystallography is appealing for its potential to provide direct
information about the interaction between an antiviral and its target, inferring how that interaction affects
a process in the viral life cycle may be difficult from such a static snapshot. Critically, because of the high
level of effort required for X-ray crystallography, it is not a feasible approach for simply screening the
potential viral targets of an unknown antiviral.
Photoaffinity cross-linking represents an alternative approach for identifying the binding site of a drug
with a known target. This technique shares strengths and weaknesses with X-ray crystallography.
Namely, photoaffinity cross-linking is useful for small molecule drugs that directly bind to and inhibit the
activity of a viral protein and does not require prior knowledge of the location of the drug binding site.512
However, inferring the mechanism of antiviral activity based on knowledge about the drug-virus protein
interaction may be difficult.
There are no alternative approaches that can determine the genetic threshold for resistance to a new
therapeutic, which is a recommended piece of data to support an Investigational New Drug (IND)
application to the FDA, prior to deployment of the therapeutic and the emergence of resistant viruses in
nature.
No alternative approaches are capable of providing information about the dose-dependence of resistance
or whether combination therapies lead to resistance less readily than individual therapies, prior to clinical
testing or post-marketing studies.
9.3.4.4 Benefits and Limitations of Alt-GoF Approaches to Both Vaccine and Therapeutic
Development
9.3.4.4.1 Vaccine/Therapeutic Development benefit #1: Testing the safety and efficacy of MCM
candidates
In addition to animal-adapted viruses, several alternative model systems could be used to test the safety
and efficacy of vaccine candidates, namely naturally susceptible hosts and transgenic animals (see Sec.
9.3.4.1.3). Transgenic mice are important in countermeasure development because they can be used to
establish that a therapy knocks down virus titers in a system with human receptors.513 However, the
predictive value of safety and efficacy data gleaned from experiments using transgenic animals is
constrained by the fact that transgenic animals do not fully recapitulate human disease pathogenesis.
510 Prabakaran P et al (2006) Structure of severe acute respiratory syndrome coronavirus receptor-binding domain complexed
with neutralizing antibody. The Journal of biological chemistry 281: 15829-15836
512 Hamouda AK et al (2014) Photoaffinity labeling of nicotinic receptors: diversity of drug binding sites! Journal of molecular
neuroscience : MN 53: 480-486

Naturally susceptible hosts of SARS or MERS are either asymptomatic or develop symptoms dissimilar
to those in humans. As a result, these “replication” models have limited utility for advanced vaccine
development. Replication models may provide easy metrics to demonstrate vaccine or drug efficacy (i.e.
reduction in viral replication), but their lack of relevant symptomology could lead to the development and
release of subpar or dangerous countermeasures. Specifically, therapeutics may cause unintended side
effects or deleterious interactions with the host immune system, which are unpredictable and may not be
observed in asymptomatic animal models.
9.3.4.4.2 Vaccine development benefit #4: Developing broad-spectrum vaccines
In addition to chimeric bat-SARS viruses generated through GoF approaches, several alternative model
systems can be used to evaluate the broad-spectrum potential of candidate MCMs. One approach involves
the use of wild type bat CoVs as challenge viruses, in lieu of chimeric bat-SARS viruses. However, the
fact that few bat CoVs can be grown in culture or in animals without the use of GoF approaches (serial
passaging or the generation of chimeric viruses) diminishes the utility of this approach.
For evaluating MCMs that target the Spike protein, the use of pseudotyped viruses represents another
alternative approach. Because Spike proteins are presented differently in the context of pseudotyped
viruses versus CoVs, results using pseudotyped viruses may not be recapitulated in the context of the wild
type virus.514 Thus, all results using pseudotyping systems must be confirmed using wild type viruses (or
chimeric CoVs, which better mimic wild type bat CoVs than pseudotyped viruses). Finally, chimeric
viruses that have been engineered to express “internal” (i.e. non-Spike) CoV proteins have been used for
testing the efficacy of therapeutics targeting non-Spike proteins.515 As with pseudotyped viruses, due to
significant differences in the course of infection between chimeric virus systems and wild type viruses,
such chimeric virus systems can be used to screen therapeutic candidates but do not replace the need to
test MCMs against the wild type virus.
approaches
9.3.5.1 Benefits to Scientific Knowledge
Serial passaging, a GoF approach that alters host range, is uniquely capable of identifying novel viral
genetic traits and factors that contribute to cross-species adaptation. Moreover, to elucidate the molecular
mechanisms underlying the role of the Spike-receptor interaction in host adaptation, testing the
phenotypic consequences of mutations in animal CoV Spike proteins in the context of a chimeric virus
generated through GoF approaches provides a higher level of certainty in the validity of the results than
similar confirmatory experiments using recombinant proteins or pseudotyped viruses. However, sequence
comparisons, an alt-GoF approach, are uniquely capable of identifying genetic traits that are associated
with mammalian adaptation across a variety of strains as well as discovering genetic markers that are
definitively associated with human adaptation. However, the causality of markers identified through
sequence analysis must be confirmed with a GoF experiment, and the utility of the comparative sequence
approach is severely compromised by the poor state of genetic surveillance for CoVs in human and
animal populations and the fact that it is limited to analysis of strains that have caused human infections.
514 Ibid.
515 Deng X et al (2014) A chimeric virus-mouse model system for evaluating the function and inhibition of papain-like
proteases of emerging coronaviruses. Journal of virology 88: 11825-11833

In addition to the approaches described above, characterizing SARS/MERS-like animal CoVs, thought to
be precursors for SARS/MERS or to have similar potential to spill over into human populations, also
provides insight into how SARS and MERS emerged from their animal reservoirs to infect humans.
However, most animal CoVs grow poorly, if at all, in standard cell culture systems. GoF approaches have
unique potential to enable the development of in vitro model systems for the study of any animal CoV in
a variety of cell types, including immortalized cell lines and relevant primary cell lines such as human
epithelial airway cells. Alternatives to GoF have significant shortcomings. Only a subset of animal CoVs
identified to date can be cultured in bat, human, or other standard cell lines, limiting the utility of using
naturally permissive cell lines for in vitro studies. While ectopic expression of permissive receptor
proteins in a common cell line has been shown to permit replication of several CoVs, this strategy is
limited to cell lines that can be readily transfected (i.e. not primary cell lines) and overexpression of the
host receptor may alter the biology of infection, limiting the relevance of results from this system.
of CoVs
Serial passaging for the selection of CoV strains with enhanced pathogenicity in animals or fitness in cell
culture, a GoF approach, is the most efficient and effective method for identifying novel genetic traits
and/or viral factors that contribute to virulence in any coronavirus strain. The alternate approaches have
several drawbacks. While screening gene knockout viruses in vitro represents a viable approach for the
discovery of novel virulence factors, this LoF approach is limited to the identification of proteins that
influence replicative fitness, only one component of virulence, and may uncover factors that attenuate
virulence for trivial reasons. The main drawback of both the GoF and LoF approaches is that insights
gleaned from model systems may not translate to human infection. To that end, comparatively analyzing
the sequences of SARS/MERS strains with varied levels of virulence can provide direct insight into
genetic traits that are associated with pathogenicity in humans. However, this approach is limited to the
study of SARS and MERS and is significantly constrained by shortcomings in the quality and availability
of existing genetic surveillance data. In addition, any hypothesis generated through comparative sequence
analysis must be experimentally confirmed. The phenotypic consequences of mutations that are
associated with enhanced virulence can be validated using GoF approaches, which are uniquely capable
of demonstrating that mutations are necessary and sufficient to enhance virulence, or LoF approaches,
which can demonstrate that mutations are necessary for enhanced virulence only. Complex, multi-genic
traits such as virulence are difficult to tease apart using solely LoF approaches because LoF provides
limited information about how proteins cooperate to enhance virulence. However, because the value of
the information gleaned from both LoF and GoF approaches depends on the relevance of artificially
manipulated viruses to nature, using both approaches to confirm the role of a particular mutation or
phenotype strengthens any conclusion.
Mouse-adapted strains of SARS, which exhibit altered host range and enhanced virulence in mice relative
to the wild type SARS virus, represent the only model system that recapitulates disease pathogenesis
observed during human infections of SARS-CoV. As existing animal models for MERS-CoV do not
replicate human disease pathology, mouse-adapted strains of MERS-CoV are expected to serve as the
sole pathogenesis model for the study of MERS-CoV infection as well. As such, animal-adapted strains
can be used to study many facets of disease pathogenesis, including the course of disease, the role of viral
and host immune factors in disease pathology, and the role of tissue tropism in disease pathology.
Alternative model systems have critical drawbacks for the study of disease pathogenesis. Because
transgenic animals do not recapitulate the features of human disease, lessons learned about pathogenesis
may not translate to humans. Most naturally susceptible hosts are asymptomatic or display dissimilar

symptoms to humans and thus cannot be used to study disease pathogenesis. While human autopsy data
are uniquely capable of providing insight into human disease pathology, limited autopsy data are
available, and the static nature of the data and the presence of co-morbidities in many SARS/MERS
patients complicate interpretation of that data.
9.3.5.2 Benefits to the development of vaccines
Live attenuated vaccines (LAVs) are being actively researched for their potential as CoV vaccine
platforms. GoF approaches that enhance virulence represent the most efficient and effective strategy for
identifying CoV virulence factors, which may be good targets for attenuation. LoF approaches are
relatively inefficient for the discovery of novel virulence factors but are critical for demonstrating that
mutagenesis or knockout of a particular virulence factor is sufficient to attenuate viral replication, the goal
of generating an LAV candidate.
LAVs are an appealing type of vaccine for CoVs for several reasons, and multiple LAV candidates for
SARS have been shown to completely protect against lethal virus challenge in mice, demonstrating the
promise of this type of vaccine for CoVs.516,517 However, one significant concern associated with LAVs is
their potential to regain virulence in people. Several other types of CoV vaccines are in development,
which do not rely on GoF approaches, and many have shown promise in animal models. Each alternative
vaccine type has strengths and weaknesses relative to LAVs, and the type or types of vaccines that will
ultimately prove to be most effective for SARS, MERS, and SARS/MERS-like coronaviruses are not yet
clear based on vaccinology research conducted to date. 518 Given the need for CoV vaccines, pursuing all
promising strategies for vaccine development in tandem, including LAVs, will ensure that an effective
vaccine is achieved in the shortest possible period of time.
GoF approaches, namely serial passaging of LAVs in cells or animals, are uniquely capable of
determining whether a candidate LAV will recover fitness/virulence upon growth in cells or animals.
Because a tendency to revert or acquire compensatory mutations that enhance fitness/virulence could
seriously compromise the safety of a live attenuated vaccine, demonstrating the genetic stability of a
candidate LAV is a critical aspect of its development.
9.3.5.3 Benefits to the development of therapeutics
As described above (Sec. 9.3.5.1.1), GoF approaches represent the most efficient and effective strategy
for identifying novel CoV virulence factors, which may serve as good therapeutic targets. However,
follow-up studies are needed to develop therapeutics that inhibit the function of that virulence factor and
to determine whether blocking its function is sufficient to reduce disease-associated pathology and/or
viral shedding. High-throughput screening of small molecule compounds for their ability to block viral
replication in vitro has generated several promising therapeutic candidates but is limited to the discovery

of therapeutics that inhibit viral replication, which is only one aspect of virulence. Several research
groups are pursuing the development of monoclonal antibodies targeting the CoV Spike protein, as mAb
binding has been shown to inhibit the ability of the virus to bind and infect cells, but mAb-based
therapeutics suffer several drawbacks relative to small molecule drugs and other types of therapeutics. As
for CoV vaccines, the type or types of therapeutics that will ultimately prove to be effective against CoVs
is not yet clear based on current research. Given the need for CoV therapeutics, pursuing all promising
strategies for therapeutic development in tandem will ensure that an effective therapeutic is achieved in
the shortest possible period of time.
Serial passaging of a virus in the presence of therapeutic to discover mutations that confer resistance, a
GoF approach, is uniquely capable of identifying the viral target of a novel therapeutic with an unknown
mechanism of action. For therapeutics with known viral targets, this information about resistance
mutations can provide foundational information to guide follow-up structural, cell biological, and
biochemical studies investigating the mechanism of action of the therapeutic. Although crystallography
and photoaffinity cross-linking can also provide insight into the antiviral mechanisms of therapeutics that
directly bind to and inhibit virus proteins, inferring mechanistic information based on static information
about the virus-antiviral complex may be difficult. Finally, the identification of host factors that are
required for antiviral activity is a critical aspect of examining therapeutics with unknown targets but
provides limited mechanistic information about therapeutics that target virus proteins.
Additionally, serial passaging of viruses in the presence of therapeutic is uniquely capable of determining
the genetic threshold for resistance development prior to deployment of the therapeutic and the emergence
of resistant strains in nature.
As both mechanism of action and selection for resistance studies to determine the genetic threshold for
resistance development are recommended components of an Investigational New Drug application to the
FDA, GoF approaches that lead to the generation of viruses that are resistant to therapeutics in
development are essential for the licensing of new therapeutics.
GoF approaches that lead to the generation of viruses that are resistant to therapeutics in development are
uniquely capable of determining the therapeutic dose that is least likely to lead to the acquisition of
antiviral resistance as well as determining whether combination therapies better prevent the emergence of
resistant viruses than individual therapies. Both types of information benefit the development of
therapeutic strategies that will be effective for a longer period of time in the field.
9.3.5.4 Benefits to the development of vaccines and therapeutics
9.3.5.4.1 Vaccine/Therapeutic development benefit #1: Testing the safety and efficacy of MCM candidates
The use of animal-adapted strains of CoVs is critical for advanced MCM development as well and
provides significant advantages over the use of alternative model systems. Though transgenic animals and
naturally susceptible hosts can be used to demonstrate that MCMs diminish viral replication, an important
proof of concept for early stage MCMs, animal-adapted strains that replicate human disease pathology
provide a much more robust system for demonstrating the safety and efficacy of MCM candidates.
Because adapted strains provoke a response from the host immune system, use of these strains can reveal

MCM side effects or adverse reactions that are not seen in asymptomatic models, an important aspect of
safety testing.
9.3.5.4.2 Vaccine development benefit #4: Developing broad-spectrum vaccines
Chimeric bat-SARS CoV strains created using GoF approaches that adapt a virus to a new host are
uniquely capable of providing reliable information about the broad-spectrum potential of CoV MCMs.
Because most bat CoV strains cannot be cultured, the use of wild type viruses cannot provide information
about whether CoV MCMs are capable of targeting a variety of SARS/MERS-like CoVs in addition to
SARS and MERS. While expressing CoV Spike proteins in the context of other viruses (i.e. pseudotyped
viruses and other chimeric virus systems) may be useful for screening MCM candidates targeting the
Spike protein, all results must be confirmed using wild type strains (or CoV chimeric strains) due to
significant differences in the behavior of chimeric viruses versus CoVs.
9.4 Introduction to GoF Research Involving Influenza Viruses
9.4.1 Overview of the landscape of GoF Research Involving Influenza Viruses
Sec. 9.4 through Sec. 9.11 describe the benefits of GoF approaches involving influenza viruses, based on
an analysis of the outcomes of GoF experiments published in the scientific literature. Our review of the
influenza virus literature included the following virus strains:
• Human seasonal strains: currently circulating and historical influenza A H1N1 and H3N2 viruses
and influenza B viruses,
• Human pandemic strains: the 1918 H1N1, 1957 H2N2, 1968 H3N2, and 2009 H1N1 viruses,
• Swine-origin strains: H3N2v and others, and
• Avian-origin strains: H5N1, H7N9, H9N2 and others.
We identified approaches involving influenza viruses that are reasonably anticipated to lead to the
following phenotypic changes:

• Altered host range,
• Evasion of existing natural or induced immunity,
• Evasion of therapeutics, and
• Evasion of vaccines in development.
Through this document, the term “therapeutics” includes drugs that directly target viruses (e.g. influenza
neuramidase inhibitors), monoclonal antibody-based therapeutics, host immune modulators, and any other
type of anti-viral therapeutic. Influenza research that is reasonably anticipated to lead to evasion of
diagnostics was not identified.
Descriptions of individual experimental approaches are provided within individual GoF phenotype
sections. Of note, passaging of influenza viruses and coronaviruses in cells is essential for any

experimental work involving live viruses, both to prepare virus stocks for experimental use and to
conduct infection experiments. This applies to alt-GoF approaches, such as characterization of wild type
viruses, as well as to GoF approaches. Because of the high mutation rates of RNA viruses, including
influenza viruses and coronaviruses, such passaging inevitably selects for higher-yield viruses.519
However, within the “enhanced virus production” phenotypic category, this analysis is restricted to those
approaches that deliberately seek to enhance virus production through serial passaging, targeted genetic
modification, or other approaches.
9.4.2 Use of attenuated strains of influenza viruses
Throughout the field of influenza research, the use of reassortant strains comprised of gene segments from
a wild type strain and an attenuated, high-yield lab-adapted strain (e.g. A/Puerto Rico 8—PR8) is
common. As described in Sec. 4.4.2.1, these strains are comprised of the HA and NA genes from a wild
type strain and the remaining six genes from PR8 (“6:2R strains”) and can be generated through reverse
genetics or classical co-infection methods. 6:2R strains can be considered GoF strains by two criteria: (1)
6:2R strains exhibit enhanced virus production relative to the parental wild type strain (albeit reduced
virus production relative to the parental lab-adapted strain), and (2) 6:2R strains exhibit enhanced
pathogenicity relative to the parental lab-adapted strain (albeit reduced pathogenicity relative to the
parental wild type strain). These strains are used for two purposes. In the context of vaccine production
and basic science research that aims to elucidate mechanisms regulating the growth of vaccine viruses
(Sec. 4.4.2.1), 6:2R strains are utilized due to their enhanced growth phenotype, which is desirable for
efficient vaccine production. Therefore, when considering enhanced viral growth, the generation and use
of 6:2R strains is considered to be a GoF approach. In contrast, in the context of research involving the
study of other GoF phenotypes associated with the HA and NA proteins, 6:2R strains are utilized due to
their attenuated phenotype, as a risk mitigation strategy. (In this context, 7:1R strains, comprised of the
HA and/or NA genes from the wild type strain and the remaining six or seven genes from PR8, may also
be used.) Specifically, researchers may perform GoF approaches, such as serial passaging of viruses in the
presence of cognate antibodies to select for antibody escape mutants, using 6:2R strains in lieu of wild
type viruses. Thus, in these studies, 6:2R strains are subjected to an additional GoF approach. These
studies include:
• Antigenic escape studies, which may lead to the generation of viruses that evade existing natural
or induced adaptive immunity,
• Emergence of antiviral resistance studies, which may lead to the generation of viruses that evade
therapeutics,
• Some approaches that may lead to the generation of viruses with enhanced pathogenicity, and
• Some approaches that may lead to the generation of viruses with altered host range in mammals
and/or enhanced transmissibility in mammals.
In each of these sections, the use of 6:2R strains in place of wild type strains is considered to be an
alternative approach for several reasons: (1) 6:2R strains are utilized due to their attenuated phenotype
relative to parental, wild type strains, (2) the use of attenuated reassortant strains has been described as an
alternative approach in the GoF debate, and (3) for a given GoF approach, the utility and limitations

associated with the use of attenuated reassortant strains are different from those associated with the use of
wild type strains and must be evaluated separately.
Several other risk mitigation strategies that involve the use of attenuated or replication-incompetent
strains may be used for select GoF approaches in lieu of wild type strains. For those GoF approaches that
may enhance pathogenicity, alter host range, or enhance transmissibility and that focus on the function of
influenza proteins other than the HA and NA, another type of risk mediation reassortant may be used.
Specifically, these reassortants comprise the HA and/or NA genes from a human seasonal flu strain, to
which the population has pre-existing immunity, and up to the remaining six or seven genes from animal
influenza strains or the 1918 H1N1 pandemic strain. As for attenuated reassortants with PR8, the use of
these strains is considered to be an alternative approach because the purpose for their use is their
attenuated phenotype and because their utility and limitations for a given GoF approach are different from
those of the wild type strains.
For select GoF studies involving highly pathogenic avian influenza (HPAI) viruses, the multi-basic
cleavage site (MBCS) of the viral surface glycoprotein HA, a major determinant of virulence, can be
removed through deletion or mutation to mitigate risk. The MBCS is thought to mediate systemic
replication and enhanced virulence in part by defining sensitivity to tissue specific proteases that are
required for activation of the HA protein during infection. As a result, HPAIΔMBCS strains do not
efficiently infect animal models and so cannot be used for in vivo studies, but these attenuated strains
can be used for select in vitro studies when cell culture media is supplemented with the appropriate
proteases.
Finally, replication-incompetent viruses can be used for the in vitro study of phenotypes underlying
pathogenicity.In these model systems, viral replication and immune evasion pathways, both of which
contribute to pathogenicity in vivo, can be assessed in cell culture lines that are engineered to stably
express an essential viral protein that is missing from the “replication-incompetent” virus strains used for
infection. For example, the replacement of the PB2 gene with a GFP-expression construct that has the
necessary flanking, non-coding and packaging sequences from the viral genome can only replicate in cell
lines that stably express exogenous PB2.520 The result is a virus that is biologically constrained to
replication in that cell line. Several replication incompetent model systems have been developed for the
study of seasonal, pandemic, and animal influenza virus gene segments.
9.4.3 Organization of the assessment of the benefits of GoF research involving influenza viruses
The following chapters (9.5 through 9.11) summarize the results of the assessment of the benefits of GoF
research involving influenza viruses. A more detailed analysis to further support the findings described in
these chapters is presented in Appendix IV Sec. 15.2 through Sec. 15.8. As the relative ability of a given
GoF (or alt-GoF) approach to address a particular scientific knowledge or public health gap often hinges
on nuanced differences between the benefits and limitations of different approaches, readers who seek an
in-depth understanding of the benefits of GoF research are directed to Appendix IV Sec. 15.
520 Ozawa M et al (2011) Replication-incompetent influenza A viruses that stably express a foreign gene. The Journal of
general virology 92: 2879-2888

9.5 Influenza viruses: Benefits of GoF Research that Enhances Virus Production
9.5.1 Summary
This section describes the benefits of GoF research that is reasonably anticipated to lead to enhanced
production of influenza viruses as the result of changes in the replication cycle or growth. Such GoF
studies were found to generate scientific knowledge, have direct applications in vaccine development, and
are likely to have economic benefits. Alt-GoF approaches that may generate similar benefits were also
identified and analyzed. At present, GoF studies resulting in enhanced influenza production have unique
and direct benefits, particularly to vaccine development and production. Chapter 9.5 provides an overview
of these benefits, including basic background and supporting information; a fully referenced and more
thorough discussion of these benefits can be found in Appendix IV Sec. 15.2.
9.5.1.1 Benefits of GoF that Enhances Virus Production to Scientific Knowledge
• GoF approaches are:

o Uniquely capable of discovering mutations that enhance the growth of vaccine viruses to
greater-than-wildtype levels, which is desirable for vaccine production and provides a
foundation for understanding the mechanistic basis of high growth of vaccine viruses,
o Uniquely capable of demonstrating that particular markers are necessary and sufficient to
enhance the growth of vaccine viruses, and
o Uniquely critical for generating information that can be translated to the vaccine production
process.
• Alternative experimental approaches are:

o Limited to the study of genetic markers and phenotypes underlying naturally high levels of
growth, and
o Limited to confirming that mutations are necessary for high growth but cannot be used to
demonstrate that mutations enhance growth, and thus cannot provide information that can be
applied to the vaccine production process.
9.5.1.2 Benefits of GoF that Enhances Virus Production to Vaccine Production

o Uniquely critical for the current capability to produce sufficient and effective vaccines for
seasonal and pandemic influenza,
o Capable of improving the quality and availability of influenza vaccines in the future by:
 Shortening production timelines for egg- and cell-based vaccines, which translates to
faster vaccine availability during a pandemic and improved vaccine match for seasonal
influenza vaccines by enabling strain selection closer to the start of flu season,
 Enabling the production of well-matched vaccines for strains that mutate to alter their
antigenicity during growth in eggs or cells, which is a unique benefit of GoF approaches.

o Because there are likely to be no regulatory barriers for incorporating genetic markers
identified through GoF research into vaccine strains, this benefit can be achieved in the
immediate to near future.
• Alternative approaches are:

o Incapable of replacing current vaccine production processes in the near-term,
o Capable of improving the quality and availability of influenza vaccines in the future through
several different mechanisms:
 Incorporating adjuvants into existing vaccines would shorten production timelines by
enabling the use of smaller quantities of antigen per vaccine dose. However, no
adjuvanted seasonal vaccines are FDA-licensed, and the development and licensing
procedures for new adjuvanted vaccines are lengthy and expensive,
 New virus-free vaccine platforms that do not rely on GoF are capable of producing strain-
specific vaccines on shorter timescales than existing egg- and cell-based production
systems. However, only one recombinant vaccine is licensed, and the development and
licensing procedures for new vaccines are lengthy and expensive,
 Developing a universal or broad-spectrum flu vaccine would obviate the need for annual
production of seasonal vaccines and for production of strain-specific vaccines in response
to the emergence of a novel pandemic strain. However, the feasibility of producing a
universal flu vaccine is unknown,
 Developing pre-pandemic vaccines would lead to faster vaccine availability during a

pandemic. However, because resources for pre-pandemic vaccine development and
stockpiling are limited, pre-pandemic vaccines function to bridge the gap between the
emergence of a novel strain and the large-scale deployment of strain-specific vaccines,
 Improving strain selection capabilities will reduce the likelihood of vaccine mismatch
due to incorrect prediction of which strains will be circulating six to nine months hence.
The realization of this benefit, which complements efforts to shorten vaccine production
timelines by addressing a different shortcoming in the existing vaccine production
process, depends on scientific advancements.
9.5.1.3 Economic Benefits of GoF that Enhances Virus Production
• Increasing the yields of vaccine viruses, using information or products derived from GoF
approaches that enhance virus production, is likely to lower the cost per vaccine dose by enabling
the production of a greater number of vaccine doses using the same materials.
GoF approaches that enhance virus production do not benefit surveillance, informing policy decisions, or
the development of therapeutics or diagnostics.
9.5.2 Overview of GoF research landscape: enhanced virus production
9.5.2.1 Generation of attenuated, high-yield candidate vaccine viruses through reassortment
Reassortment between a wild type strain and an attenuated, high-yield vaccine backbone strain generates
a “Candidate Vaccine Virus” (CVV), which comprises the HA and NA genes from the wild type strain

and the remaining six “internal genes” from the vaccine backbone strain. CVVs are attenuated and exhibit
higher levels of growth relative to the parental, wild type virus. CVVs may be generated through classical
reassortment methods, which involve co-infection of eggs or cells with the wild type strain and the
vaccine backbone strain followed by antibody-based selection for viruses with the correct surface
antigens, or through reverse genetics.521 CVVs serve as the basis of vaccine strains that are used for the
production of influenza vaccines in eggs or cells. Additionally, in the context of academic research,
comparing the sequences of CVVs with varied growth properties enables the identification of mutations
that are associated with high yield.
9.5.2.2 Serial passaging of viruses in eggs or cells
Serial passaging of viruses in eggs or cells selects for higher-yield viruses. This approach is currently
used for the production of influenza vaccines in eggs or cells as well as for basic science research on the
mechanisms underlying high growth of influenza viruses. For vaccine production, manufacturers serially
passage CVVs in eggs or cells to generate high-yield vaccine seed strains that can be used for large-scale
production of vaccines. In the context of scientific research, serial passaging of viruses in eggs or cells
followed by sequencing of the emergent higher-yield viruses enables the identification of mutations that
are sufficient to enhance the growth of viruses. Subsequently, mutant viruses are subjected to antigenic
characterization using the hemagglutinin inhibition (HAI) assay or other assays to identify which
mutations confer high growth without changing the antigenicity of the strain. For research purposes, this
approach is most commonly carried out using vaccine backbone strains and CVVs but may also be carried
out using wild type strains.
9.5.2.3 Forward genetic screen to identify mutations that confer high growth to viruses
Forward genetic screens, which involve random mutagenesis of viruses followed by limited passaging to
select for mutants with high growth properties, enable the identification of mutations that confer high
growth to viruses. Forward genetic screens involving vaccine backbone strains and CVVs lead to the
identification of mutations that are sufficient to enhance the yields of vaccine viruses. Subsequently,
mutant viruses are subjected to antigenic characterization using the hemagglutinin inhibition (HAI) assay
or other assays to determine which mutations confer high growth without altering the antigenicity of the
strain.
9.5.2.4 Targeted mutagenesis of viruses to introduce mutations that are associated with high growth
Targeted mutagenesis of viruses to introduce mutations that are associated with high growth, followed by
characterization of virus yields relative to the parental virus, demonstrates that a mutation or set of
mutations is necessary and sufficient to confer high growth. Subsequently, antigenic characterization
assays are performed to confirm that the mutations have not altered the antigenicity of the virus, and the
mutant strain is subjected to several rounds of passaging in eggs or cells to ensure that it is genetically
stable – that is, that it does not acquire additional mutations that alter its antigenicity upon further growth.
This knowledge provides a foundation for follow-up studies investigating the mechanistic basis of the
high-growth phenotype (e.g. the use of cell biological assays, biochemical assays, and other assays to
explore how the mutation enhances growth). Notably, these mutations may have been discovered through
a GoF approach, such as serial passaging or a forward genetic screen, or through an alt-GoF approach,
such as comparative analysis of wild type sequences.
backbone strain.

Finally, it should be noted that experimental approaches involving targeted genetic modification of the
viral polymerase complex of avian viruses to render it more “human-like” (through site-directed
mutagenesis or reassortment between human and avian viruses) is also likely to enhance virus replication.
However, as the primary goal of those studies is to gain insight into the mechanisms underlying
adaptation of avian viruses to mammals, those studies are discussed in the “enhanced transmission in
mammals” section.
9.5.3 Identification of potential benefits and limitations of GoF approaches
In this section, the potential benefits of GoF research that enhances virus production in each benefit
category listed in the NSABB Framework are discussed.
9.5.3.1 Scientific Knowledge Benefits
The mechanistic basis of high growth of vaccine viruses in eggs or cells is not well understood. Current
strategies for producing vaccine viruses do not consistently produce high-yield strains, which are needed
for efficient vaccine production. In addition, further boosting the growth properties of all vaccine strains
has potential to increase the efficiency of the existing vaccine production process. The genetic and
phenotypic traits that promote the growth of vaccine strains are largely unknown.
GoF approaches that enhance virus production have the potential to address this scientific knowledge gap
by providing insight into the genetic and mechanistic basis of the enhanced growth phenotype.
Specifically, serial passaging of viruses in eggs/cells and forward genetic screens followed by selection of
high-growth mutants enable the identification of mutations that are sufficient to confer higher-than-
wildtype levels of growth to any virus strain, though results may not translate to other virus strains.
Comparative analysis of the sequences of CVVs with varying growth properties can also lead to the
identification of mutations that are associated with naturally high levels of growth and may be more likely
to uncover determinants of high growth that are conserved across multiple strains. However, comparative
sequence analysis is unlikely to uncover genetic markers associated with greater-than-wild type levels of
growth because it is limited to analysis of existing isolates. Finally, targeted mutagenesis can be used to
demonstrate that a particular mutation or set of mutations is necessary and sufficient to enhance the
growth of a vaccine virus strain, across multiple strain contexts. Collectively, these approaches provides a
foundation for follow-up structural, biochemical, and cell biological assays investigating the phenotypic
consequences of the mutations to gain insight into the mechanisms underlying the enhanced growth
phenotype.
9.5.3.2 Surveillance
All other GoF approaches are focused on identifying mutations that confer high growth to vaccine viruses
(either candidate vaccine viruses or vaccine backbone strains). Because these viruses have no correlate in
nature, this information does not inform the interpretation of genetic surveillance data from animals or
humans.
9.5.3.3 Development and Production of Vaccines
9.5.3.3.1 Background – Shortcomings in Existing Influenza Vaccine Production Processes
GoF approaches that enhance virus production have the potential to benefit the development of influenza
vaccines, which are strain-specific. Due to antigenic drift of circulating seasonal influenza viruses, the
strain composition of influenza vaccines must be updated annually, and the CDC’s Advisory Committee
on Immunization Practices recommends annual influenza vaccination for all people ages six months and

older.522 Currently, over 99% of seasonal influenza vaccines used in the US are produced in eggs or cells,
and the same systems and facilities would be used to produce pandemic vaccine in response to the
emergence of a novel pandemic strain of influenza.523,524 For production of both seasonal and pandemic
influenza vaccines, the vaccine production cycle spans six to eight months.525,526
Though the influenza vaccine development and production systems are well-established, interviews with
stakeholders in the influenza research and public health communities highlighted that the lengthy
production timelines for existing egg- and cell-based vaccines critically limit the mitigating impact of
influenza vaccination on the morbidity and mortality associated with influenza outbreaks. Lengthy
vaccine production timelines impact the quality and availability of seasonal and pandemic flu vaccines
differently. In the context of seasonal flu epidemics, existing production timelines necessitate strain
selection nine months in advance of the peak of the target flu season.527 As a result, one or more vaccine
strains are often imperfectly matched to circulating strains, which reduces the efficacy of the vaccine.528
In the context of pandemics, vaccines are simply unavailable to protect the public until at least six months
into the outbreak.529,530
9.5.3.3.2 Potential Benefits and Limitations of GoF Approaches to Current Influenza Vaccine Production
GoF approaches are core aspects of the current process for producing influenza vaccines. To be suitable
for large-scale manufacturing of vaccine virus, a selected field isolate must be attenuated and its growth
in eggs/cells must be enhanced. This enhancement is achieved through the use of two different GoF
approaches: (1) a CVV is created through reassortment between the field isolate and an attenuated, high-
yield vaccine backbone strain and (2) the CVV is serially passaged in eggs or cells to increase its yield.
Collectively, these GoF approaches increase HA antigen yield by yield at least 12-fold relative to the
cognate wildtype strain. 531 (It should be noted that manufacturers report production increases in terms of
HA antigen yield rather than viral titer because the FDA requires that a certain quantity of HA antigen be
present in each vaccine dose. Increases in viral titer correlate with increases in HA antigen yields.)532,533
The use of high-growth reassortant viruses generated through GoF methods enables the production of
over 170 million doses of seasonal influenza vaccine annually and would enable the production of a
similar number of doses of pandemic vaccine six to eight months after the emergence of a novel
pandemic strain.534
522 CDC’s Advisory Committee on Immunization Practices (ACIP) Recommends Universal Annual Influenza Vaccination.
http://www.cdc.gov/media/pressrel/2010/r100224.htm. Last Update Accessed September 15, 2015.
523 Dowling B. Protein Sciences' N.Y. Factory Licensed For Flu Vaccine Production. http://www.courant.com/business/hc-
protein-sciences-pearl-river-approval-20150513-story.html. Last Update 13 May 2015. Accessed 14 September 2015.
524 CDC. What You Should Know for the 2015-2016 Influenza Season. http://www.cdc.gov/flu/about/season/flu-season-2015-
2016.htm. Last Update Accessed September 15, 2015.
525 (2015e) Current practices in influenza vaccine production. Interview with Industry or Federal Government Representative
with Expertise in Influenza Vaccine Production.
526 Stöhr K (2013a) Influenza vaccine production. In Textbook of Influenza, 2nd Edition, 2nd Edition edn, pp 352-370.
527 Ibid.
528 (2015n) Influenza Vaccine Strain Selection. Interview with Academic Researcher or Federal Government Representative
Involved in the Annual Strain Selection Process for Seasonal Influenza Vaccines.
529 Borse RH et al (2013) Effects of vaccine program against pandemic influenza A(H1N1) virus, United States, 2009-2010.
531 Ibid.
532 Food and Drug Administration. Annex 5: Vaccination Development and Production - Draft
http://www.hsdl.org/?view&did=459937. Last Update Accessed September 15, 2015.

9.5.3.3.3 Potential Benefits and Limitations of GoF Approaches to Future Influenza Vaccine Production
The insights gleaned from GoF approaches that enhance virus production also have the potential to
improve vaccine production practices in the future through two distinct mechanisms: (1) shortening
vaccine production timelines, and (2) improving the match between the virus strains used as the basis of
vaccine strains and the strains that are circulating during flu season (referred to as “vaccine match”).
First, GoF approaches have the potential to shorten vaccine production timelines by increasing the yields
of vaccine viruses, which govern the rate at which vaccines are produced and thus serves as a key
determinant of the time needed for egg- and cell-based vaccine production. GoF approaches can increase
CVV yields in two ways: (1) through the direct use of higher-yield CVVs and (2) through the
incorporation of genetic markers that confer high growth into existing CVVs using targeted mutagenesis.
Shortening vaccine production timelines improves seasonal and pandemic influenza vaccines through
different mechanisms. In the context of pandemics, faster vaccine production translates to vaccine
availability earlier during the pandemic. In the context of seasonal flu epidemics, the ability to produce
vaccines in a shorter period of time enables strain selection closer to the start of flu season, increasing the
likelihood that the vaccine strains will match the circulating strains during peak flu season. Ultimately,
increasing the availability of vaccines during a pandemic and increasing the efficacy of vaccines during
seasonal flu epidemics will reduce human morbidity and save lives.535
One key constraint on the benefits afforded by improvements to CVV yields is the limited production
capacity of eggs and cells. Current egg-based vaccine production systems are at or near maximal levels of
production, suggesting that the benefits of GoF research are largely limited to improving the growth of
“poor” CVVs.536 However, because many CVVs based on zoonotic viruses and seasonal H3N2 viruses
grow poorly in eggs, simply improving their production would significantly benefit public health.537,538 In
contrast, the production capacities of cell-based systems have not yet plateaued, thus GoF research that
improves CVV yields has the potential to benefit production of vaccines for all influenza sub-types using
cell-based systems.539 Importantly, because minor modifications to existing CVVs are unlikely to require
FDA approval for use in vaccine production, these benefits can be realized in the immediate future.540
Second, the insights derived from GoF research can improve vaccine match for vaccines based on strains
that tend to mutate upon growth in eggs or cells, which may lead to antigenic changes and poor vaccine
match. In particular, H3N2 strains often acquire antigenicity-altering mutations upon growth in eggs,
which is especially concerning given that H3N2 strains tend to cause more severe disease than H1N1
537 (2015o) Candidate vaccine virus development. Interviews with Influenza Researchers Involved in Candidate Vaccine Virus
Development.
540 Ibid.

strains.541,542,543,544 Mutations that enhance the growth of these strains without altering antigenicity,
identified through GoF studies, can be incorporated into CVVs to enable the production of vaccines that
match the antigenicity of selected strains.
9.5.3.4 Therapeutics and Diagnostics
Information about mutations that confer high growth to vaccine viruses or about mutations that rescue the
growth of antiviral resistant strains is not relevant to the development of therapeutics.
The process of developing influenza diagnostics is well-established.545,546 GoF research that leads to the
identification of genetic markers that confer GoF phenotypes, including enhanced virus production, does
not inform diagnostic development.
9.5.3.5 Informing policy decisions
Similarly, information about mutations that confer high growth to vaccine viruses does not inform the
analysis of genetic surveillance data, so this information does not benefit policy decisions about public
health preparedness.
Increasing the yields of vaccine viruses, using information or products derived from GoF approaches that
enhance virus production, is likely to lower the cost per vaccine dose by enabling the production of a
greater number of vaccine doses using the same materials. However, the economic benefits of
enhancements to vaccine virus yields to vaccine production were not explored in detail in this report.
9.5.4 Identification of the potential benefits and limitations of alt-GoF that provide similar potential
benefits to the GoF being examined
public health benefits as GoF approaches but through different mechanisms.
9.5.4.1 Potential Benefits and Limitations of alt-GoF Approaches to Scientific Knowledge
Several alternative experimental approaches can provide insight into the genetic and mechanistic basis of
high growth of vaccine viruses, which complement GoF approaches that lead to the identification of
541 (2015o) Candidate vaccine virus development. Interviews with Influenza Researchers Involved in Candidate Vaccine Virus
Development.
542 Barman S et al (2015) Egg-adaptive mutations in H3N2v vaccine virus enhance egg-based production without loss of
antigenicity or immunogenicity. Vaccine 33: 3186-3192
543 Huang SSH et al (2011) Comparative Analyses of Pandemic H1N1 and Seasonal H1N1, H3N2, and Influenza B Infections
Depict Distinct Clinical Pictures in Ferrets. PLoS ONE 6: e27512
544 Kaji M et al (2003) Differences in clinical features between influenza A H1N1, A H3N2, and B in adult patients.
Respirology (Carlton, Vic) 8: 231-233
545 New diagnostics for novel influenza viruses are typically real-time PCR assays which include two or three diagnostic
targets. The influenza M gene is used as a marker for influenza A, the HA gene is used for sub-typing, and the NA gene may
also be included. Developing of a new diagnostic assay simply requires designing new primers and probes for a virus of
interest, which requires that the sequences of the M, HA, and NA genes are available.
546 (2015a) Interview with U.S. CDC (Centers for Disease Control) representative.

genetic traits that enhance virus growth. First, sequence comparison of wildtype strains with varied
growth properties can lead to the identification of mutations that are associated with naturally high levels
of growth. Of note, because of the importance of genetic context on multi-genic traits such as fitness,
mutations that confer high growth to wildtype strains may not confer high growth to vaccine strains (i.e.
reassortants that include the HA and NA from the field isolate and the remaining six genes from a vaccine
backbone strain). Additionally, this approach depends on the existence of high-growth strains in nature
and cannot identify mutations that confer exceptional yields.
Genetic screens to identify mutations that reduce growth (i.e. loss-of-function, or LoF) can lead to the
discovery of mutations that are necessary for growth. A major limitation of this approach is that it may
uncover mutations that reduce growth for “trivial,” reasons, i.e. that modulate critical aspects of virus
function that are necessary for viability but do not directly contribute to high growth). An additional
drawback is that it is much less efficient than its GoF counterpart because mutants must be screened for
reduced growth (versus selection for high growth through passaging). Finally, the utility of the
information gleaned from LoF screens also depends on the existence of high-growth strains in nature.
LoF approaches may also be used to confirm that a particular amino acid residue (discovered through
GoF or alt-GoF approaches) is necessary for high growth. However, the marker may not be sufficient to
enhance growth if introduced into a different strain, limiting the utility of this result for vaccine
production.
9.5.4.2 Potential Benefits and Limitations of alt-GoF Approaches to Vaccine Production
9.5.4.2.1 Potential Benefits and Limitations of Alt-GoF Approaches to Current Influenza Vaccine
Production
High-yield, attenuated vaccine viruses generated through GoF approaches are currently used for
production of egg- and cell-based vaccines. The use of strains with wild type growth properties represents
one alternative to the use of high-yield vaccine viruses. This could involve the direct use of wild type
strains or the use of novel reassortant strains that are attenuated but exhibit wild type levels of virus
production. Because most influenza strains grow poorly in eggs/cells, the concentration of virus antigen in
eggs/cells infected with strains with wild type growth properties would be so low that existing
manufacturing processes would likely fail to purify antigen that meets FDA standards, resulting in no
vaccine produced. Alternatively, a wild type isolate with exceptional growth properties could be used to
produce the same number of doses over a longer period of time or to produce a smaller number of doses
over the same period of time. For example, use of a wild type isolate the grows four times as well as an
average strain would either lengthen the vaccine production timeline to more than one year or would
reduce the number of doses produced two- to three-fold. Additionally, because wild type viruses with
exceptional yields and appropriate antigenic properties are unlikely to be available, this scenario would
likely result in the use of a poorly matched strain, leading to the production of a less effective vaccine.
Additionally, neither alternative (i.e. use of wild type strains or use of novel reassortants with wild type
growth properties) can be implemented immediately. Large-scale production using wild type isolates for
the purpose of producing inactivated vaccines would pose significant risks to vaccine manufacturers prior
to the inactivation step, presumably requiring the construction of new manufacturing facilities capable of
virus production under higher biocontainment conditions. Of note, field isolates cannot be used as a basis
for live vaccines due to their pathogenicity. The alternative, use of attenuated vaccine viruses with wild
type growth properties, would necessitate the development, and perhaps subsequent FDA licensing, of
novel vaccine backbone strains that attenuate but do not confer high growth to reassortant viruses.

As described above, production of virus-based vaccines in eggs/cells necessitates passaging of the
antigenic strain of interest to produce enough stock virus to infect eggs/cells for large-scale
manufacturing, which inevitably selects for higher-yield viruses due to the high mutation rate of influenza
viruses.547 If this passaging were considered to be a GoF approach, in addition to the approaches
described above that deliberately enhance the yields of vaccine viruses, then completely avoiding
manipulations that are reasonably expected to enhance virus production precludes production of egg- and
cell-based influenza vaccines. In that case, virus-free vaccine platforms, such as recombinant or DNA-
based vaccines, represent an alternative to egg- and cell-based flu vaccines.548,549,550 However, only one
recombinant flu vaccine is commercially available and is only approved for use in people 18 years of age
and older. This vaccine represented just 50,000 of more than 140 million doses administered during the
2014 – 2015 flu season.551,552 Although other recombinant vaccines are in late stages of development,
given the long and expensive product development cycle for new influenza vaccines – spanning 8 – 12
years and costing 300 million to one billion dollars including research, clinical development, and
registration with the FDA – alternative, virus-free flu vaccine platforms are not a viable replacement for
egg- and cell-based vaccines in the immediate future.553
9.5.4.2.2 Potential Benefits and Limitations of Alt-GoF Approaches to Future Influenza Vaccine
Production
Several alternative scientific and technical innovations have the potential to benefit vaccine production in
the future. Of note, some of these innovations can improve the production of both seasonal and pandemic
influenza vaccines, whereas others can only improve production of seasonal or pandemic vaccines. These
differences reflect the fact that seasonal flu vaccines are produced annually in advance of flu season,
whereas pandemic vaccines are produced in response to the emergence of a novel pandemic strain.
An alternative approach for improving vaccine virus yields without enhancing the inherent growth
properties of CVVs is through modulation of the host cells that are used to produce virus. Specifically,
identification of host genes that suppress viral growth provides a basis for development of specialized
knockout cell lines that permit higher virus yields.554 This approach has potential to benefit the production
of seasonal and pandemic flu vaccines but has been tested on a limited number of strains. No modified
cell lines are currently FDA-approved for vaccine production, and only one cell-based vaccine that could
potentially make use of this technology is licensed in the U.S.555 Cell lines must undergo extensive testing
in order to be FDA-approved for influenza vaccine production prior to their commercial use, which will
delay realization of this benefit.556,557 Finally, because this approach increases viral titer, it is not less risky
than GoF approaches than enhance the growth properties of vaccine viruses.
549 Kim JH, Jacob J (2009) DNA vaccines against influenza viruses. Current topics in microbiology and immunology 333: 197-
210
550 Bright R. Review of New Vaccine Platforms and Influenza Vaccine Pipeline.
http://www.who.int/influenza_vaccines_plan/resources/bright.pdf. Last Update Accessed September 15, 2015.
552 Protein Sciences. Flublok. http://www.proteinsciences.com/FVAC.htm. Last Update Accessed September 15, 2015.
554 Hamamoto I et al (2013) High yield production of influenza virus in Madin Darby canine kidney (MDCK) cells with stable
knockdown of IRF7. PloS one 8: e59892
555 TABLE. Influenza vaccines — United States, 2015–16 influenza season.
http://www.cdc.gov/flu/protect/vaccine/vaccines.htm. Last Update Accessed September 14, 2015.
557 FDA. Guidance for Industry: Characterization and Qualification of Cell Substrates and Other Biological Materials Used in
the Production of Viral Vaccines for Infectious Disease Indications

An adjuvant is a substance that is added to a vaccine to boost the body’s immune response to the vaccine,
and including an adjuvant in a vaccine may enable the use of a smaller quantity of antigen to induce the
same level of protection (“dose sparing”). 558 Thus, incorporating adjuvants into existing egg- and cell-
based vaccines represents a different strategy for shortening production timelines, by enabling production
of the same number of doses over a shorter period of time. Most licensed vaccines in the U.S. are not
adjuvanted – one seasonal vaccine containing adjuvants was recently approved for use in people aged 65
and older, and one licensed pandemic influenza vaccine contains adjuvants.559,560,561,562 Nonetheless, use of
adjuvants to improve the immunogenicity of seasonal influenza vaccines is an active area of research. The
major barrier to realization of this benefit is that existing vaccines that are re-formulated with adjuvant are
considered new drugs by the FDA and as such must undergo the standard licensure pathway for
unadjuvanted vaccines, which will delay their widespread availability due to the time needed to generate
the needed safety and efficacy data.563,564,565 This approach has potential to benefit the production of
seasonal and pandemic flu vaccines.
Developing new vaccine platforms with faster production timelines represents a third alternative approach
for shortening the time needed for production of strain-specific vaccines. Recombinant vaccines, which
are virus-free vaccines comprised of recombinant influenza proteins produced in insect cells or other
protein expression systems such as plants, represent the most developed and promising approach.566,567
Although only one recombinant vaccine is currently FDA-licensed, several other recombinant vaccines
are in late stages of development, and experts in the influenza vaccine field expect the production and use
of this type of vaccine to increase over the next several decades. 568,569 However, as mentioned above, the
time needed for completion of clinical trials and licensing delays the ability of this technology to impact
http://www.fda.gov/downloads/biologicsbloodvaccines/guidancecomplianceregulatoryinformation/guidances/vaccines/ucm2
02439.pdf. Last Update Accessed September 15, 2015.
558 CDC. Vaccine Adjuvants. http://www.cdc.gov/vaccinesafety/concerns/adjuvants.html. Last Update Accessed September 15,
2015.
559 Ibid.
560 Influenza A (H5N1) Virus Monovalent Vaccine, Adjuvanted.
http://www.fda.gov/BiologicsBloodVaccines/Vaccines/ApprovedProducts/ucm376289.htm. Last Update Accessed
September 15, 2015.
561 FDA. FDA approves first seasonal influenza vaccine containing an adjuvant. FDA News Release.
http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm474295.htm. Last Update November 24, 2015.
Accessed November 28, 2015.
562 Novartis. FLUAD® (MF59®-Adjuvanted Influenza Vaccine) Fact Sheet.
https://www.novartis.com/sites/www.novartis.com/files/Fluad_Fact_Sheet.pdf. Last Update Accessed September 15, 2015.
563 Montomoli E et al (2011) Current adjuvants and new perspectives in vaccine formulation. Expert Rev Vaccines 10: 1053-
1061
564 Food and Drug Administration. Vaccine Product Approval Process.
http://www.fda.gov/BiologicsBloodVaccines/DevelopmentApprovalProcess/BiologicsLicenseApplicationsBLAProcess/ucm
133096.htm. Last Update 24 August 2015. Accessed 14 September 2015.
565 Gruber M. Regulatory Pathways Supporting Development and Approval of Vaccines Formulated with Novel Adjuvant:
Regulatory Considerations and Challenges.
http://www.fda.gov/downloads/EmergencyPreparedness/MedicalCountermeasures/UCM292045.pdf. Last Update 2012.
Accessed 14 September 2015.

influenza vaccination systems in the US in the near term (i.e. within the next few years). 570 Virus-free
vaccine platforms can be used for the production of seasonal and pandemic influenza vaccines.
A universal or broad-spectrum flu vaccine would obviate the need for yearly production of strain-specific
vaccines as well as the need to produce a strain-specific vaccine in response to the emergence of a novel
pandemic strain. Such a vaccine could be administered in advance of a pandemic, generating pre-existing
immunity in the population, or could be stockpiled and immediately deployed following the start of a
pandemic. However, universal or broader-spectrum vaccines are still in early stages of development and
represent an extremely challenging prospect given the high mutability of influenza viruses.
Development of pre-pandemic vaccines against circulating zoonotic influenza strains with pandemic
potential would also lead to faster vaccine availability during a pandemic caused by a closely related
strain. Developing pre-pandemic CVVs and carrying out clinical trials would shorten vaccine production
timelines, and stockpiling bulk antigen would allow for near-immediate deployment of vaccine following
emergence of a pandemic strain. In addition, manufacturers’ experience with production of the vaccine
would likely streamline subsequent large-scale production during the pandemic.571 Although the pre-
pandemic vaccine strain is unlikely to exactly match the strain that emerges to cause a pandemic, use of
adjuvants and prime-boost regimens broaden the protection that can be achieved using a strain-specific
vaccine, such that pre-pandemic vaccines are highly likely to provide some level of protection against
infection with a similar strain.572,573,574,575,576 The benefit of developing pre-pandemic vaccines is
constrained by the fact that resources for the development and stockpiling of pre-pandemic vaccines are
limited. Resource limitations necessitate targeted, risk-based investments in pre-pandemic vaccine
development, which are informed by GoF approaches that enhance the transmissibility and virulence of
influenza viruses, discussed in detail in Sec 9.6.3.3.577
Shortening vaccine production timelines (through GoF or alt-GoF approaches) represents one strategy for
improving the match between seasonal flu vaccines and circulating strains, by enabling the selection of
vaccine strains closer to the start of flu season. However, as long as vaccine strains must be selected in
advance of flu season, there remains the possibility of vaccine mismatch due to incorrect prediction of
which strains will be dominant during the target flu season. Therefore, improving strain selection
capabilities represents a completely different mechanism for improving the efficacy of seasonal influenza
vaccines. As discussed in detail in Sec 9.8.5.3.1, both GoF approaches that lead to evasion of existing
natural or induced adaptive immunity and alt-GoF approaches have potential to improve strain selection
capabilities.
572 Ibid.
573 (2015d) Influenza Vaccines. Interviews with Public Health Professionals Involved in Preventing and Responding to
Influenza Outbreaks.
574 Smith GE et al (2013) Development of influenza H7N9 virus like particle (VLP) vaccine: homologous A/Anhui/1/2013
(H7N9) protection and heterologous A/chicken/Jalisco/CPA1/2012 (H7N3) cross-protection in vaccinated mice challenged
with H7N9 virus. Vaccine 31: 4305-4313
575 Middleton D et al (2009) Evaluation of vaccines for H5N1 influenza virus in ferrets reveals the potential for protective
single-shot immunization. Journal of virology 83: 7770-7778
576 Khurana S et al (2010) Vaccines with MF59 adjuvant expand the antibody repertoire to target protective sites of pandemic
avian H5N1 influenza virus. Sci Transl Med 2: 15ra15
577 Cox NJ et al (2014) Pandemic preparedness and the Influenza Risk Assessment Tool (IRAT). Current topics in
microbiology and immunology 385: 119-136

approaches
In this section, the potential benefits of GoF research that enhances virus production relative to alt-GoF
approaches are discussed, in each benefit category that GoF approaches can address
GoF approaches are uniquely capable of discovering mutations that enhance the growth of vaccine
viruses to greater-than-wildtype levels, which is important for the efficient production of egg- and cell-
based influenza vaccines. In addition, GoF approaches are uniquely capable of demonstrating that
particular markers are necessary and sufficient to enhance the growth of vaccine viruses, which is
essential for translation of information about high-growth markers to the vaccine production process.
Together, this information provides a foundation for follow-up studies investigating the mechanistic basis
of high virus yields. Alternative approaches have significant limitations for the study of high virus yields,
relative to GoF approaches. Comparative sequence analysis of wildtype viruses is limited to the study of
genetic markers and phenotypes underlying naturally high levels of growth, which provides a limited
range of information relative to GoF approaches. LoF approaches are much less efficient than GoF
approaches, and mutations that attenuate growth may not confer high growth if introduced into a new
strain. Neither alternative approach is capable of generating information that can be applied to vaccine
productions.
9.5.5.2 Benefits to Vaccine Production
9.5.5.2.1 Benefits to Current Production of Influenza Vaccines
GoF approaches to enhance the growth of attenuated vaccine strains are a uniquely critical component
of the current ability to produce sufficient and effective vaccines for seasonal and pandemic influenza.
The use of strains with field-like growth properties in lieu of high growth reassortants generated using
GoF approaches would result in the production of no vaccines, the production of a lesser quantity of
vaccines that poorly match circulating strains, or the production of vaccines that poorly match circulating
strains over an extended time period. Furthermore, using field strains would require the construction of
new manufacturing facilities capable of virus production under higher biocontainment conditions, and
using attenuated strains with wildtype levels of growth would require the development and possible FDA
licensure of new vaccine backbone strains that are attenuated but do not confer high growth to reassortant
viruses, delaying the implementation of either alternative approach. Recombinant vaccines and other
virus-free vaccine platforms represent a promising approach for future influenza vaccine production, but
the one recombinant vaccine that is currently licensed represents less than 1% of seasonal influenza
vaccines administered annually, and lengthy regulatory processes will delay the availability of additional
virus-free vaccines in the future.
9.5.5.2.2 Benefits to Future Production of Pandemic Influenza Vaccines
Both GoF approaches to improve CVV yields and alternative approaches have potential to reduce the
time lag between the emergence of a novel pandemic strain in human populations and the widespread
availability of a vaccine, thus reducing human morbidity and mortality during an influenza pandemic.
GoF approaches to improve the yields of vaccine viruses are uniquely capable of achieving this benefit
in the immediate to near term because use of this information capitalizes on existing infrastructure and
faces no regulatory barriers to translation. Adjuvanted vaccines and virus-free vaccines also have shorter
production timelines than existing egg- and cell-based vaccines, but the widespread availability of both
types of vaccines will be delayed due to time needed for vaccine development and the generation of

safety and efficacy data that is required for FDA licensure of new flu vaccines. Universal flu vaccines are
not a viable option for protection against pandemic influenza in the near future, and whether the
development of a universal or broad-spectrum vaccine is possible is unknown. The development of pre-
pandemic vaccines represents a promising strategy due to their ability to provide broad-spectrum
protection when adjuvanted. However, because resources limit the scope of the USG’s investment in pre-
pandemic vaccines, these vaccines will serve to bridge the gap between the emergence of a novel strain
and widespread availability of vaccines and must be complemented by innovations to shorten vaccine
production timelines.
9.5.5.2.3 Benefits to Future Production of Seasonal Influenza Vaccines
Both GoF approaches and alt-GoF approaches have potential to improve the match between seasonal
influenza vaccines and strains that are circulating during flu season, thus improving vaccine efficacy and
decreasing human morbidity and mortality associated with seasonal flu epidemics. This benefit can be
achieved through several different mechanisms. One strategy is to improve the production of strains that
mutate to alter their antigenicity upon growth in eggs or cells, which results in the production of vaccines
that are poorly matched to the selected strains. GoF approaches are uniquely capable of generating high-
yield, genetically stable CVVs that do not acquire antigenicity-altering mutations during passage in eggs
or cells. A different strategy is to shorten the time needed to produce influenza vaccines, which enables
strain selection closer to the start of flu season. GoF approaches to improve the yields of CVVs are
uniquely capable of achieving this benefit in the immediate to near term relative to alternative approaches
for shortening vaccine production timelines, for the reasons described above. Finally, a completely
different mechanism for increasing the likelihood of vaccine match is to improve strain selection
capabilities, which can be achieved through both GoF and alt-GoF approaches (see Sec 9.8.5.3.1).
Though promising, benefits in this area rely on scientific advancements and the expansion of influenza
surveillance networks, thus both the extent and timescales of the benefits are uncertain. Notably, because
this approach addresses different underlying gaps in existing vaccine development and production
processes, research in this area has potential to complement the benefits that can be achieved by GoF
research that shortens vaccine production timelines by increasing CVV yields.
9.6 Influenza viruses: Benefits of GoF research that enhances mammalian adaptation and
transmissibility
9.6.1 Summary
This section describes the benefits of GoF research that is reasonably anticipated to enhance the
infectivity and transmissibility of influenza viruses in representative animal models. Such GoF studies
were found to generate scientific knowledge and to inform surveillance of circulating animal influenza
viruses, which has downstream impacts on decision-making about USG investments in pandemic
preparedness initiatives such as pre-pandemic vaccine development. Alt-GoF approaches that may
generate similar benefits were also identified and analyzed. At present, GoF studies resulting in enhanced
infectivity and transmissibility in mammals have unique benefits to scientific knowledge, surveillance,
and pandemic preparedness, though full realization of GoF benefits to public health requires significant
scientific advancements. Sec. 9.6 provides an overview of these benefits, including basic background and
supporting information; a fully referenced and more thorough discussion of these benefits can be found in
Sec 15.3.

9.6.1.1 Benefits of GoF that Enhances Mammalian Adaptation and Transmissibility to Scientific
Knowledge

o Uniquely capable of proactively determining whether any animal influenza virus can evolve
the capacity for airborne transmission in mammals.
o Uniquely capable of providing in-depth information about the evolution of mammalian

adaptation/transmissibility in any animal influenza strain, and of determining the order of
acquisition of changes that are necessary and sufficient to enhance infectivity/transmissibility
in mammals. However, laboratory results may not translate to adaptation of animal influenza
viruses to humans in nature.
o Uniquely capable of discovering novel genetic and phenotypic traits underlying mammalian
adaptation and transmissibility in any virus strain, and of establishing a causal link between a
particular trait and enhanced mammalian adaptation/transmissibility. However, results from
cell culture or animal studies may not translate to human disease.
• Alt-GoF approaches are:

o Limited to determining whether existing animal influenza viruses can efficiently transmit
between mammals.
o Uniquely capable of providing direct insight into the evolutionary mechanisms underlying
adaptation of animal influenza viruses to humans, but cannot provide direct insight into the
evolution of human transmissibility, as animal influenza viruses that efficiently transmit
between humans do not exist in nature.
o Uniquely capable of identifying novel genetic traits associated with adaptation of animal
influenza viruses to humans, but cannot identify traits associated with human transmissibility,
as animal influenza viruses that efficiently transmit between humans do not exist in nature.
o Capable of identifying novel genetic traits in known phenotypes underlying mammalian

adaptation and transmissibility using in vitro, virus free systems, but results may not be
recapitulated in the context of the full virus.
9.6.1.2 Benefits of GoF that Enhances Mammalian Adaptation and Transmissibility to surveillance
• GoF approaches:
o Provide a foundation for the development of rapid assays for phenotypes underlying
mammalian adaptation and transmissibility, which have potential to increase the quantity and
timeliness of phenotypic information about animal flu viruses detected through surveillance.
However, the success of this approach is subject to significant advancements in the state of
knowledge about mechanisms underlying mammalian adaptation and transmissibility.
o Are uniquely capable of strengthening the predictive value of molecular markers for
mammalian adaptation and transmissibility, which can be used to infer phenotype from
sequence. Use of molecular markers in lieu of or to corroborate phenotypic testing results
could improve the quality, timeliness, and quantity of phenotypic information about animal
flu viruses detected through surveillance. However, the success of this approach is subject to
significant advancements in the state of knowledge the mechanistic basis of mammalian
adaptation and transmissibility, and predictions must be experimentally validated.

o Are critical for improving computational models for predicting phenotypes underlying
mammalian adaptation and transmissibility based on sequence, which could improve the
quantity and timeliness of phenotypic information about animal flu viruses detected through
surveillance. However, the success of this approach is subject to significant advancements in
the state of knowledge about the mechanistic basis of mammalian adaptation and
transmissibility, and predictions must be experimentally validated.
o Have significant limitations for advancing the development of rapid assays for phenotypes
underlying mammalian adaptation and transmissibility and for strengthening the predictive
value of molecular markers for mammalian adaptation and transmissibility.
o Are also critical for improving computational models for predicting phenotypes underlying
mammalian adaptation and transmissibility based on sequence, but through the generation of
different types of data that complement data generated through GoF approaches.
o Phenotypic assays for mammalian adaptation and transmissibility are uniquely capable of
providing direct information about each complex phenotype under controlled conditions, but
results may be delayed relative to the publication of viral sequences or, in the future, the
generation of data about underlying phenotypes through rapid assays.
9.6.1.3 Benefits of GoF that Enhances Mammalian Adaptation and Transmissibility to decision-
making in public health policy
• GoF approaches:
o Molecular markers for mammalian adaptation and transmissibility, which can be generated
and validated using GoF approaches, moderately influence pandemic risk assessments of
circulating animal influenza viruses, relative to other types of data that are considered.
Pandemic risk assessments guide downstream decisions about investments in pre-pandemic
vaccines, which will increase vaccine availability during a pandemic if a similar strain
emerges to cause a pandemic.
o Molecular marker data plays a relatively important role when novel influenza viruses first
emerge in human populations, when epidemiological data are scarce and virological data are
not yet available. The ability to conduct a rapid risk assessment using molecular marker data
can provide a three to four week head start on vaccine production.
o Molecular marker data can guide selection of particular viruses to use as the basis of pre-
pandemic vaccines, when multiple viruses have similar epidemiological and virologic
characteristics.
o Epidemiological data are the most influential data in a pandemic risk assessment, but
transmissibility can be difficult to assess in human populations, and epidemiological data may
be scarce when novel viruses first emerge in human populations.
o Virologic data strongly influence pandemic risk assessments, but the generation of virological
data may be delayed relative to the publication of sequencing data when novel viruses emerge
abroad, due to shipping delays.

o Other types of data, such as ecological data, also contribute to pandemic risk assessments but
completely molecular marker data (GoF) by evaluating completely different aspects of
pandemic potential.
9.6.2 Overview of GoF research landscape: enhanced infectivity and transmissibility in

representative animal models
9.6.2.1 Serial passaging of viruses in mammalian cells or animals
Serial passaging of viruses in mammalian cells in laboratory animals selects for viruses with enhanced
growth in cells or enhanced infectivity to animals, respectively. This type of serial passaging experiment
involves “forced” passaging, meaning that the experimenter directly transfers infected material, in the
form of cell culture supernatant or homogenates of infected tissue, to the subsequent cell culture dish or
animal. Forced serial passaging is carried out for two purposes: (1) to identify mutations that arise during
adaptation of animal influenza viruses (i.e. avian and swine viruses) to mammals, which provides a
foundation for follow-up studies investigating the evolutionary mechanisms driving adaptation to
mammalian hosts and the mechanistic basis of mammalian adaptation, and (2) to develop an mouse model
for the study of a particular virus.
9.6.2.2 Serial passaging of viruses in mammalian cells or animals with selection for transmission
Serial passaging of viruses in animals with selection for transmission leads to the generation of viruses
with enhanced transmissibility in mammals. This type of serial passaging experiment can involve
selection for contact transmission, during which the primary (directly inoculated) and secondary hosts are
co-housed, or for airborne transmission, during which the primary and secondary hosts are separately
housed in special isolator cages that prevent direct contact between animals but allow for air exchange
between cages. These studies seek to identify mutations that are sufficient to enhance transmissibility,
which provides a foundation for follow-up studies that investigate the mechanistic basis of
transmissibility in mammals.
9.6.2.3 Forward genetic screen to identify genetic traits that enhance the fitness/transmissibility of
viruses in mammals
Forward genetic screens involve random mutagenesis of genetic regions predicted to contribute to
fitness/transmissibility or comprehensive reassortment of parental gene segments from two viruses,
followed by characterization of the fitness or transmissibility of mutants in appropriate mammalian model
systems to select for mutant viruses with enhanced fitness/transmissibility. Sequencing emergent viruses
enables the identification of mutations or gene segments that enhance the fitness/transmissibility of
viruses, which provides a foundation for follow-up studies that investigate the mechanistic basis of
9.6.2.4 Targeted genetic modification of viruses to introduce traits that are expected to enhance
fitness/transmissibility in mammals
Targeted genetic modification of viruses, namely site-directed mutagenesis and/or reassortment, to

introduce genetic traits that are expected to enhance the fitness/transmissibility of viruses followed by
characterization of the fitness or transmissibility of mutants in appropriate mammalian model systems
may lead to the generation of viruses with enhanced fitness/transmissibility in mammals. This approach is
performed for two purposes: (1) to determine whether a previously characterized underlying genetic or
phenotypic trait, such as a preference for binding to α2,6 sialic acid receptors, contributes to the complex
phenotypes of mammalian adaptation or transmissibility, and (2) to confirm that a particular mutation or

gene segment is necessary and sufficient to enhance the fitness/transmissibility of viruses in appropriate
model systems. Notably, genetic traits that are associated with mammalian adaptation/transmissibility
may be discovered through GoF approaches or alt-GoF approaches. As above, this information provides a
foundation for follow-up studies investigating the mechanistic basis of mammalian adaptation and
transmissibility.
9.6.3 Identification of the potential benefits and limitations of GoF approaches
In this section, the potential benefits of GoF research that enhances mammalian adaptation and
transmissibility in each benefit category listed in the NSABB Framework are discussed.
9.6.3.1 Benefits and limitations of GoF approaches to Scientific Knowledge
GoF approaches have potential to benefit several aspects of scientific knowledge about the ability of
animal influenza viruses to adapt to efficiently infect and transmit between humans. In this section, the
ability of GoF approaches to address three key outstanding questions related to influenza virus adaptation
and transmission in humans is evaluated:
• Can animal influenza viruses become transmissible between humans?
• How do animal influenza viruses adapt to and become transmissible between humans? What
selective pressures drive adaptation and the evolution of efficient transmissibility, and what is the
order of acquisition of new genetic/phenotypic traits that are needed for
adaptation/transmissibility?
• What is the mechanistic basis of adaptation and transmission in humans? What viral factors are
involved, and what phenotypic changes must occur in order for an animal influenza virus to adapt
to efficiently infect, cause disease, and transmit in mammals?
Viral fitness and transmissibility are complex phenotypes that arise through the cumulative effects of
multiple underlying phenotypes, such as specificity for a particular type of cell surface receptor and the
ability to replicate within a particular temperature range. Because the property of transmissibility depends
on phenotypes underlying both adaptation and transmission and because similar experimental approaches
are used to study both complex phenotypes, GoF experiments that enhance adaptation and transmissibility
are discussed together. Several phenotypes have been shown to be associated with mammalian adaptation
and transmissibility. However, considerable gaps in knowledge remain about the molecular basis of each
phenotype and the role of each phenotype in adaptation/transmissibility, and as-yet-undiscovered viral
factors and phenotypic changes are likely to contribute to the acquisition of efficient transmissibility in
mammals. Furthermore, the potential for animal influenza strains to evolve efficient transmissibility in
humans is not understood.
9.6.3.1.1 Scientific Knowledge Gap #1: Can animal influenza viruses become transmissible between
humans?
Several GoF approaches can lead to the generation of transmissible viruses, including deliberate genetic
modification of viruses and serial passaging of viruses in animals with selection for transmission.
Collectively, these approaches definitively demonstrate that a virus can acquire the capacity to transmit
between laboratory animals in an experimental setting. Notably, this approach can be applied to strains
that have not yet caused infections in human populations as well as strains that have caused human

infections but do not yet efficiently transmit in humans. The key limitations of this approach are that
observations in animal models may not translate to humans and that the adaptive changes observed in the
laboratory may not be possible in nature.
9.6.3.1.2 Scientific Knowledge Gap #2: How do animal influenza viruses adapt to and become
transmissible in humans?
Serial passaging of animal influenza viruses in appropriate animal models to select for mammalian
adaptation and transmission, a GoF approach, provides insight into the mechanisms underlying adaptation
to mammals and the evolution of transmissibility. Sequencing of isolates at multiple stages of passaging
enables determination of the order and rate of acquisition of adaptive traits, and follow-up studies
elucidate how those genetic and phenotypic changes influence other viral phenotypes. Comparing the
sequences and phenotypes of viral isolates from different tissues, different time points during the course
of infection, and between the primary (directly inoculated) and the secondary hosts can provide additional
insight into the tissue-dependence of adaptation, the rate of intra- and inter-host adaptation, and the
selection pressures and viral population dynamics during transmission, respectively. Notably, the adaptive
changes that occur in the lab environment under forced selection may not be relevant or possible during
natural evolution, may not mimic adaptation and transmission in humans, and may selectively represent
the evolutionary course possible for the limited number of viruses studied.
Serial passaging provides information about the genetic traits that are associated with the acquisition of
enhanced fitness and transmissibility in mammals. However, to confirm which of these changes are
necessary and sufficient to enhance fitness and transmissibility, targeted mutagenesis must be used to re-
introduce mutations into parental strains followed by characterization of the infectivity/transmissibility of
mutant strains. Targeted mutagenesis also enables determination of how the order of acquisition of
genetic changes influences other viral phenotypes, such as replicative fitness, which has implications for
the likelihood that these traits can arise in nature.
9.6.3.1.3 Scientific Knowledge Gap #3: What are the genetic and phenotypic traits that result in
adaptation and transmission in humans?
Several GoF approaches can be used to discover the genetic and phenotypic markers underlying
mammalian adaptation and transmission of animal influenza viruses include:
• Targeted genetic modification to introduce novel genetic changes that are expected to contribute
to adaptation and transmission in mammals by either site-directed mutagenesis or targeted
reassortment (often between animal and human seasonal strains),
• Forward genetic screens involving random mutagenesis or comprehensive reassortment followed

by selection for mammalian infectivity, transmissibility, or underlying phenotypes, and
• Serial passaging in appropriate animal models or mammalian cells to select for mammalian
adaptive or transmissible traits.
Collectively, these approaches enable the identification of genetic changes that are sufficient to confer
enhanced fitness in cell culture model systems or infectivity and transmissibility in animal models, which
provides a foundation for follow-up biochemical, cell biological, and structural studies that elucidate
associated phenotypic changes. Serial passaging has the potential to uncover novel genetic and phenotypic
markers that contribute to adaptation/ transmissibility. In contrast, because forward genetic screens
involving random mutagenesis typically focus on regions that are suspected or known to play a role in
phenotypes underlying adaptation/transmissibility, this approach can discover novel genetic markers for

adaptation/transmissibility only. The targeted genetic modification approach is limited to the investigation
of genetic traits and underlying phenotypes that are suspected to contribute to adaptation/transmissibility
(e.g. determining whether altering sialic acid receptor binding specificity contributes to transmissibility).
Targeted genetic modification is also used to confirm that particular mutations or gene segments are
necessary and sufficient to enhance infectivity or transmissibility in mammals. The use of in vitro model
systems is limited to the investigation of phenotypes underlying adaptation and transmissibility, such as
replicative fitness and sialic acid receptor specificity. Moreover, the results derived from these studies
may not be recapitulated in the complex environmental pressures encountered in a host. The relevance of
both in vitro and in vivo approaches depends on whether mechanisms underlying adaptation to cell culture
and animal models are representative of those in humans, and results gleaned from the study of one or a
few strains may not be recapitulated in different genetic contexts.
9.6.3.2 Benefits and Limitations of GoF approaches to Surveillance
This section collectively evaluates the benefits of GoF research that enhances mammalian adaptation,
transmissibility, or virulence to surveillance, as the strategies for monitoring the evolution of all three
properties in circulating animal influenza viruses and the potential benefits of GoF approaches are similar.
GoF approaches that lead to the identification of genetic and phenotypic traits underlying mammalian
adaptation, transmissibility between mammals, and virulence have the potential to inform the
interpretation of wildlife, agricultural animal, and public health surveillance data. Specifically, GoF data
has potential to improve three practices for evaluating the infectivity, transmissibility, and virulence of
surveillance isolates: (1) inspecting sequences for the presence of molecular markers for mammalian
adaptation, transmissibility, and virulence, (2) developing rapid assays for phenotypes underlying
mammalian adaptation, transmissibility, and virulence, and (3) developing computational models for
predicting underlying phenotypes. Information about the infectivity, transmissibility, and virulence of
circulating animal influenza viruses is one aspect of evaluating their risk to human populations, which
informs downstream decision-making related to public health preparedness for novel influenza outbreaks.
The contribution of GoF data to pandemic risk assessments is discussed in detail in Sec. 9.6.3.3.2, below.
9.6.3.2.1 Introduction to influenza virus surveillance: current practices and limitations
Influenza surveillance is conducted in human and animal populations, including agricultural animals,
companion animals, and wildlife. Collectively, the goal of this surveillance is to monitor the evolution of
circulating animal influenza viruses, in order to detect the emergence viruses that pose a risk of emerging
in human populations to cause a pandemic. Resources can then be dedicated to mitigating the risk factors
associated with virus emergence, for example through community-level interventions at the animal-
human interface, and to preparing for a potential emergence event, for example through the development
of pre-pandemic vaccines.578 Analysis of the phenotypic properties of individual surveillance isolates is a
key aspect of assessing their pandemic potential. This section focuses on GoF benefits to that surveillance
effort.
The WHO Global Influenza Surveillance and Response System (GISRS) serves as a central repository for
data about animal influenza infections in humans. GISRS is a two-tiered surveillance and public health
laboratory system.579,580 A global network of National Influenza Centres (NICs) collect clinical specimens
in their countries, perform preliminary analyses such as viral isolation and sub-typing, and forward
578 Ibid.
579 (2015p) Interview with Centers for Disease Control and Prevention representative.
580 WHO. Global Influenza Surveillance and Response System (GISRS). http://www.who.int/influenza/gisrs_laboratory/en/.
Last Update Accessed December 7, 2015.

specimens with suspect animal influenza infections to one of six WHO Collaborating Centres (WHOCCs)
for further characterization.581
Multiple virus properties contribute to the likelihood that the virus will adapt to efficiently transmit in
human populations and the potential consequences of that event, including whether the virus is adapted
(or poised to adapt) to efficiently infect and transmit between humans and viral virulence. These
properties can be directly measured in the laboratory or can be inferred from the genetic sequence based
on the presence of molecular markers that have been linked to those phenotypes through previous
research. In practice, due to caveats associated with both strategies, both are utilized. Two other
approaches are in development but are not yet used in public health practice. The first involves the use of
rapid assays to measure phenotypes underlying mammalian adaptation, transmissibility, and virulence
(i.e. versus evaluating the complex phenotype through animal experiments). The second involves
computational modeling to predict phenotype from genotype, which incorporates experimental data about
mutations that give rise to phenotypic changes, structural data, and other types of data. This section
evaluates how GoF approaches can benefit surveillance by improving strategies for evaluating
mammalian adaptation, transmissibility, and virulence: the use of molecular markers, the use of rapid
phenotypic assays, and the use of computational models. First, the utility and limitations of traditional
methods for laboratory evaluation of the infectivity, transmissibility, and virulence of viruses are
evaluated. This information motivates the need for development of additional approaches that can provide
information about these virus properties.
The pathogenicity and the ability of an animal influenza virus to infect and transmit in mammals is
typically evaluated in ferrets.582 The strength of these assays is that they directly measure the complex
properties of mammalian adaptation, transmissibility, and virulence. However, multiple shortcomings are
associated with reliance on these assays. First, these assays are unable to assess when viruses have
acquired underlying properties that are necessary but not sufficient to enhance infectivity, transmissibility,
or virulence. Such knowledge about partial adaptation is of interest for pandemic risk assessments.
Second, these assays require the use of surveillance isolates, which limits the number of viruses that can
be subjected to phenotypic characterization.583 Third, transmission and virulence testing in animals
requires technical expertise and must be conducted under BSL-3 conditions, limiting the conduct of these
assays to the six WHOCCs.584 This restriction is problematic when logistical, political, and regulatory
factors delay the shipment of virus samples from NICs and other field diagnostic laboratories to
WHOCCs, thereby delaying the generation of phenotypic data.585,586
For the reasons listed above, the CDC has incorporated the use of molecular markers for phenotypes of
concern into the pandemic risk assessment process, to complement virologic data. Because the
phenotypes of mammalian adaptation, transmissibility, and virulence are complex, arising from the
interplay between multiple underlying phenotypes, this strategy involves inspecting sequences for
markers that are casually linked to underlying phenotypes (e.g. altered sialic acid receptor binding
specificity). Because a constellation of amino acid changes is needed for an animal virus to evolve to
efficiently infect, transmit, and cause disease in people, molecular markers are considered collectively to
determine the overall risk associated with a virus. Importantly, this process assumes that the complex
581 WHOCCs include the U.S. Centers for Disease Control in Atlanta, GA and St. Jude Children’s Research Hospital in
Memphis, TN.
582 (2015c) Interviews with influenza researchers and government representatives involved in pandemic risk assessments.
585 Ibid.
586 WHO (2011) Pandemic influenza preparedness framework for the sharing of influenza viruses and access to vaccines and
other benefits.

phenotypes of mammalian adaptation, transmissibility, and virulence can accrue in a step-wise fashion,
such that “partially adapted” viruses can persist in nature.
Influenza research experts agree that the state of this science does not enable accurate and reliable
prediction of phenotype from genotype for complex phenotypes such as mammalian adaptation,
transmissibility, and virulence. Multiple sources of scientific uncertainty limit current capabilities, which
can be broadly grouped into two categories: (1) uncertainties related to the phenotypes underlying
adaptation, transmissibility, and virulence, and (2) uncertainties related to the genetic traits that alter
underlying phenotypes.
Uncertainties related to phenotypes underlying mammalian adaptation, transmissibility, and virulence:
1. Weak linkage between underlying phenotypes and adaptation/transmissibility/virulence – that is,

uncertainty in whether particular underlying phenotypes, such as altered sialic acid receptor
binding specificity, are necessary for complex phenotypes, such as mammalian adaptation across
many different virus strains.
2. Lack of knowledge about how underlying phenotypes interact to alter adaptation, transmissibility,
and virulence (i.e. how to integrate the presence of multiple markers to appropriately determine
overall risk).
3. Lack of knowledge about whether complex phenotypes can slowly accrue (i.e. whether partially
adapted viruses can persist in nature) or whether the acquisition of efficient infectivity,
transmissibility, and enhanced virulence in mammals is an “all-or-none” phenomenon.
Uncertainties related to the genetic traits that alter underlying phenotypes
1. Inability to predict whether a particular amino acid substitution will have similar phenotypic
consequences in new genetic contexts.
2. Lack of knowledge about whether different amino acid substitutions at a particular amino acid
position will have similar phenotypic consequences as known mutations.
3. Lack of knowledge about the mutational landscape that permits evolution of a complex
phenotype – e.g. how many different sets of mutations enable the acquisition of airborne
transmissibility? This knowledge gap influences whether the absence of a known marker is
meaningful; if the mutational landscape is poorly understood, the “negative” strain could contain
as-yet-undiscovered markers.
Collectively, these sources of uncertainty significantly compromise the predictive value of molecular
markers for mammalian adaptation, transmissibility, and virulence.
Given the shortcomings associated with phenotypic assays and molecular marker data, the use of
computational methods for sequence-based predictions of phenotypes underlying mammalian adaptation,
transmissibility, and virulence has also been proposed. Although a variety of computational methods have
shown promise for predicting phenotype from genotype, for those “known” phenotypes associated with
adaptation/transmissibility, the accuracy of their predictions remains largely unknown.587,588
588 Russell CA et al (2014) Improving pandemic influenza risk assessment. Elife 3: e03883

GoF approaches have potential to address shortcomings associated with the use of virological data,
molecular markers, and computational methods to evaluate the infectivity, transmissibility, and virulence
of animal influenza viruses in mammals, representing three different strategies for improving upon the
status quo. The value of each strategy and the utility and limitations of GoF approaches for improving
each strategy are discussed below.
9.6.3.2.2 Analysis of GoF approaches that support the development of rapid phenotypic assays
Strengths and weaknesses of using rapid phenotypic assays to inform pandemic risk assessments
Rapid assays to measure phenotypes underlying mammalian adaptation, transmissibility, and virulence
could be performed in lieu of traditional evaluation of these complex phenotypes using ferrets. The
development of rapid phenotypic assays holds promise for improving analysis of surveillance data for
several reasons. First, the use of assays that are higher throughput than ferret testing will enable the
phenotypic characterization of a larger number of viruses. Second, rapid phenotypic assays that require
less technical expertise than ferret experiments are better suited for NICs, which would shorten the time
lag between the initial detection and phenotypic characterization of a given virus. Thus, taken together,
the development of rapid phenotypic assays has the potential to expand the quantity and the timeliness of
phenotypic characterization data available for pandemic risk assessments. However, these assays will
need to be carried out under BSL-3 conditions, which will limit the number of diagnostic laboratories that
will be able to conduct the assays. (The majority of NICs do not have BSL-3 capabilities, though the
number of NICs with BSL-3 capabilities or with access to BSL-3 labs has increased since 2005.)
In order for rapid phenotypic assays to be useful as proxies for mammalian adaptation, transmissibility,
and virulence, the measured phenotype must be strongly linked to adaptation/transmissibility/virulence
across many strain contexts. Additionally, interpretation of the results requires knowledge about how
individual phenotypes contribute to overall pandemic risk, which relies on an understanding of how
underlying phenotypes synergize to shape complex phenotypes. Gaps in scientific knowledge related to
these two questions constrain the development and use of rapid phenotypic assays, as described above. As
discussed in detail in Sec. 9.6.3.1.3, GoF approaches can provide insight into these scientific questions.
The relevant findings are summarized below.
Summary – benefits of GoF approaches
GoF approaches represent the most efficient and effective approach for identifying novel phenotypic traits
underlying mammalian adaptation, transmissibility, and virulence. Furthermore, targeted genetic
modification of viruses to introduce genetic traits that alter underlying phenotypes is uniquely capable of
demonstrating that a particular phenotype is causally linked to enhanced
infectivity/transmissibility/virulence in mammals across multiple virus contexts. Additionally, the ability
to alter phenotypes individually and in combination (i.e. through incorporation of varying sets of
mutations) provides insight into how multiple underlying phenotypes interact to enhance infectivity,
transmissibility, or virulence in mammals. This approach can also determine how an “intermediate” level
of adaptation/transmissibility/virulence, i.e. acquisition of some but not all phenotypic traits that are
required for viruses to efficiently infect, cause disease, and transmit in mammals, affects viral fitness,
which may provide insight into whether such partially adapted strains can persist in nature. However, the
major caveat associated with GoF approaches is that results gleaned from laboratory studies involving
animal models may not translate to human disease in nature.

9.6.3.2.3 Analysis of GoF approaches that support the use of molecular markers to evaluate the risk
posed by circulating animal influenza viruses
Strengths and weaknesses of using molecular marker data to inform pandemic risk assessments
The use of molecular marker data to evaluate the pandemic potential of animal influenza viruses has
potential to improve the quantity and timeliness of phenotypic information about circulating animal
influenza viruses. An increasing number of NICs and other diagnostic laboratories in developing
countries have sequencing capabilities, and stakeholders involved in animal influenza surveillance stated
that viral genetic sequence data is currently the fastest and more reliable data generated by diagnostic labs
in areas where viruses of concern are circulating.589 Given the time needed for sample shipment to
WHOCCs and ferret testing, the ability to assess the phenotypic properties of viruses based on sequence
data can provide information before traditional phenotypic assays. Currently, most genetic surveillance
data is generated by sequencing of viruses at WHOCCs.590 Therefore, full realization of the benefits that
can be derived from the use of molecular marker data will require an expansion of sequencing capabilities
at diagnostic laboratories that comprise the “base” of the influenza surveillance system.
As described above, the current utility of molecular markers to the interpretation of genetic surveillance
data is constrained by multiple sources of scientific uncertainty. Additionally, as knowledge about the
phenotypes underlying mammalian adaptation, transmissibility, and virulence is incomplete, the
discovery of additional molecular markers associated with novel underlying phenotypes would broaden
the utility of this approach. As discussed in detail in Sec. 9.6.3.1.3, GoF approaches can provide insight
into these scientific questions. The relevant findings are summarized below.
Summary – benefits of GoF approaches
GoF approaches represent the most efficient and effective approach for identifying novel genetic and
phenotypic traits underlying mammalian adaptation, transmissibility, and virulence. Furthermore, targeted
genetic modification of viruses to introduce genetic traits associated with mammalian
adaptation/transmissibility/virulence is uniquely capable of establishing a causal link between a particular
genetic or phenotypic trait and mammalian adaptation, transmissibility, or enhanced virulence across
multiple strain contexts. In addition, GoF approaches, namely forward genetic screens, are uniquely
capable of systematically exploring alternative mutational pathways for altering an underlying phenotype
(e.g. changing sialic acid receptor binding specificity) in the context of whole virus. Finally, GoF
approaches are also uniquely capable of providing definitive information about how multiple phenotypes
synergize to promote mammalian adaptation, efficient transmissibility, and virulence. The major caveat
associated with GoF approaches is that results gleaned from laboratory studies involving animal models
may not translate to human disease in nature.
9.6.3.2.4 Analysis of GoF approaches that improve predictive models
Strengths and weaknesses of using computational models to inform pandemic risk assessments
As the use of computational models to predict phenotypes underlying mammalian adaptation,

transmissibility, and virulence capitalizes on (and depends on) the availability of sequence data, the
strengths and limitations of this approach are similar to those described above for the use of molecular
marker data.
590 Ibid.

Existing computational models cannot reliably predict phenotypes underlying mammalian adaptation,
transmissibility, and virulence based on sequence information. Additional experimental data is needed to
appropriately parameterize models, and experiments must be conducted to validate the phenotypic
predictions of models. GoF approaches can generate data that improves the accuracy of existing models.
Summary - benefits of GoF approaches
A variety of experimental data are needed to improve the accuracy of existing models, including data
about mutations that do and do not give rise to phenotypic changes of interest. This data is critical for
building models that can account for the context dependence of genetic changes in influenza biology. GoF
approaches (targeted mutagenesis and forward genetic screens) are uniquely capable of generating these
data in the context of the full virus, although in vitro, virus free approaches can also be used. Finally,
model predictions must be validated experimentally, and results feedback to improve model accuracy.
GoF approaches are uniquely capable of validation model predictions in the context of the full virus.
9.6.3.3 Benefits and Limitations of GoF approaches to Inform Policy Decisions
GoF approaches that lead to the identification of molecular markers for mammalian adaptation and
transmissibility between mammals contribute to assessments of the pandemic risk posed by circulating
animal influenza viruses, which are based on genetic surveillance data and several other types of data
(e.g. epidemiologic data, phenotypic data, etc.). These assessments inform policy decisions related to
public health preparedness for novel influenza outbreaks, including whether to develop pre-pandemic
vaccines. Additionally, the demonstration that avian influenza viruses can evolve the capacity for
more efficient transmission in mammals may, in and of itself, stimulate interest and investment
in pandemic preparedness initiatives. This section evaluates the potential benefits of GoF approaches
in both areas.
9.6.3.3.1 Benefits of “proof of principle” GoF research that demonstrates the capacity of a virus to
evolve more efficient transmissibility in representative animal models
Researchers have suggested that the “proof of principle” demonstration that an animal influenza virus can
evolve the capacity for airborne transmission in a laboratory setting, as a blunt indicator of the pandemic
potential of the virus, could inform government interest and investment in pandemic preparedness
initiatives. However, pandemic preparedness activities at the U.S. CDC and ASPR, including BARDA,
did not change in the wake of the laboratory demonstrations that H5N1 and H9N2 could evolve the ability
to transmit via the airborne route between ferrets in 2012 and 2009, respectively, suggesting that this is
not a real benefit.591,592,593 USG representatives involved in pandemic preparedness indicated that the
response to the demonstration that an animal virus that has not yet caused human infections can evolve
the capacity for airborne transmission would also be minimal, due to the lack of certainty about whether
laboratory results translate to humans in nature.594 If the virus were known or suspected to be circulating
in animal populations in the US, enhanced surveillance might be undertaken to better understand the
prevalence and geographic distribution of the virus in nature. However, the result would be highly
unlikely to trigger investments in pre-pandemic vaccine development.
591 Imai M et al (2012) Experimental adaptation of an influenza H5 HA confers respiratory droplet transmission to a reassortant
H5 HA/H1N1 virus in ferrets. Nature 486: 420-428
592 Herfst S et al (2012) Airborne transmission of influenza A/H5N1 virus between ferrets. Science 336: 1534-1541
593 (2015k) Interviews with CDC, ASPR, and BARDA representatives.
594 (2015j) Interviews with CDC and BARDA representatives.

9.6.3.3.2 Benefits of GoF research that informs pandemic risk assessments
The second mechanism through which GoF approaches can benefit pandemic preparedness planning is
through pandemic risk assessments, downstream of GoF benefits to surveillance. As discussed in Sec.
9.6.3.2, GoF approaches have potential to benefit virological surveillance (i.e. by supporting the
development of rapid phenotypic assays) as well as genetic surveillance (i.e. by strengthening the
predictive value of molecular markers for phenotypic properties of concern and by improving
computational models for predicting phenotype from genotype). The use of molecular markers for
phenotypic properties of concern is currently incorporated into the risk assessment process, as described
in detail below. As neither rapid assays nor robust computational models for relevant phenotypes exist,
how results from notional future assays/models would be considered in risk assessments is uncertain.
Thus, the potential benefits of rapid phenotypic assays or computational models to pandemic risk
assessments is not formally evaluated in this section, but a discussion of how results from either could
contribute to the risk assessment process is provided at the end of the section.
This section analyzes the value of using molecular marker data relative to other types of data that are
considered in the pandemic risk assessment process, which provides an “upper bound” to the public
health benefits that can be achieved through GoF improvements to surveillance. First, to provide context
for this analysis, current strategies for pandemic risk assessments are reviewed, and shortcomings in
existing processes are highlighted.
Background – pandemic risk assessment and strategies for decision-making about investments in
pandemic preparedness
The U.S. government undertakes influenza pandemic preparedness activities that aim to bolster U.S.
capabilities for rapid detection of novel influenza events and to limit the spread of disease, death, and
potential societal impacts if/when the next influenza pandemic occurs.595 In particular, the development of
pre-pandemic vaccines is a key aspect of pandemic preparedness because influenza vaccination is the
primary public health strategy during outbreaks.596 As resources for pandemic preparedness efforts are
limited, a major challenge is determining how resources for the development of pre-pandemic vaccines
and other preparedness activities should be allocated. To that end, the CDC, in collaboration with subject
matter experts in influenza virology, diagnosis, epidemiology, ecology, and laboratory research in animal
and human influenza, developed a framework for assessing the relative risk posed by emerging influenza
viruses and an accompanying tool – the Influenza Risk Assessment Tool (IRAT). Those results then
inform prioritization of resources for preparedness efforts directed at particular strains/sub-types (e.g.
vaccine development).
The IRAT provides a formal method for evaluating the relative risk posed by different emerging influenza
strains (e.g. H5N1 versus H7N9).597,598 This method is based on SME input about risk elements that
govern the likelihood that a particular strain will adapt to efficiently transmit in human populations and
the expected public health consequences of that emergence event. These risk elements can be broadly
grouped into four categories:
596 Ampofo WK et al (2013b) Strengthening the influenza vaccine virus selection and development process: outcome of the 2nd
WHO Informal Consultation for Improving Influenza Vaccine Virus Selection held at the Centre International de
Conferences (CICG) Geneva, Switzerland, 7 to 9 December 2011. Vaccine 31: 3209-3221
598 Trock SC et al (2012) Development of an influenza virologic risk assessment tool. Avian diseases 56: 1058-1061

• Elements relating to the properties of the virus (e.g. transmissibility and virulence),
• Elements relating to the attributes of host populations (e.g. the degree of pre-existing immunity),
• Elements relating to epidemiological, and
• Elements relating to ecological factors (e.g. the extent of human infections and the prevalence and
geographic distribution of the virus in animal populations).
Selected elements will be described in more detail below. Risk elements pertaining to the properties of the
virus are informed by virological data (e.g. transmission studies in ferrets) and by genomic data, including
molecular marker data (e.g. whether molecular markers associated with enhanced transmissibility in
ferrets are present in the viral genetic sequence). Individual risk elements are weighted, based on SME
input about their relative contribution to the likelihood and expected consequences of emergence of
particular strains, and all elements are considered collectively to determine an overall risk score. Notably,
although not all policy decisions related to pandemic preparedness rely on formal risk assessments, the
same factors are considered when informally evaluating the risks posed by emerging influenza viruses.
Potential Benefits of GoF to pandemic risk assessments: utility and limitations of using molecular marker
data
GoF approaches have potential to improve the accuracy, timeliness, and quantity of phenotypic
information generated by inspecting sequences for the presence of molecular markers for mammalian
adaptation, transmissibility, and virulence, as described in detail in Sec. 9.6.3.2.3. This section focuses on
the utility and limitations of molecular marker data to the pandemic risk assessment process.
Molecular marker data
Molecular markers for phenotypes underlying mammalian adaptation, transmissibility, and virulence are
considered as part of the “genomic variation” risk element (which also incorporates consideration of
reassortment). As described above, these analyses complement results from laboratory-based phenotypic
assays. The major strength of this analysis is that sequence data are now the fastest, most reliable data
produced at NICs and other field laboratories where animal influenza viruses of concern are circulating,
enabling evaluation of viruses prior to the generation of virological data in the laboratory.599 However, the
predictive value of molecular markers is compromised by significant sources of scientific uncertainty, as
described above. Because of these uncertainties, molecular marker data contributes moderately to the risk
assessment, relative to other factors. For example, in the three-virus relative risk assessment referenced
above, findings related to epidemiology risk elements were about six-fold more important than findings in
the genomic variation risk element.
9.6.3.3.3 Public health impacts of pandemic risk assessments
Pandemic risk assessments are carried out to help prioritize resources for investments in pre-pandemic
vaccine development. Risk assessments may also guide investments in other pandemic preparedness
initiatives, such as testing the efficacy of antivirals against high-risk viruses. GoF approaches aid
decision-making downstream of pandemic risk assessments insofar as GoF-derived data contributes to the
pandemic risk assessment process. Additionally, GoF approaches can be used to select particular viruses
to be used as the basis of pre-pandemic vaccine strains.
599 (2015h) Interview with CDC Representative.

Pre-pandemic vaccine development
Because existing influenza vaccines are strain-specific, pre-pandemic vaccines are developed to target
particular groups of high-risk strains. Depending on the overall level of risk associated with a particular
virus, the U.S. government will fund development of a pre-pandemic vaccine through various stages of
the vaccine production pipeline. Each of the following steps requires an escalating expenditure of
resources: CVV development, conduct of pre-clinical vaccine studies in animals, manufacture of clinical
trial lots of vaccine, conduct of human clinical trials, stockpiling of vaccine, and priming the population
against the novel influenza virus. Collectively, these investments will increase the availability of vaccines
during a pandemic by shortening vaccine production timelines.600 Although the pre-pandemic vaccine
strain is unlikely to exactly match the strain that emerges to cause a pandemic, use of adjuvants and
prime-boost regimens broaden the protection that can be achieved using a strain-specific vaccine, such
that pre-pandemic vaccines are highly likely to provide some level of protection against infection with a
similar strain.601,602 Notably, resources limit the scope of the USG’s investment in pre-pandemic vaccines,
highlighting the need for strategies to prioritize vaccine development for the many influenza viruses
circulating in nature that have spilled over into human populations.
GoF data may play a role in the decision to develop a CVV for an animal influenza virus by contributing
to the pandemic risk assessment process, particularly when new viruses first emerge in human
populations and sequence data are available before other types of virologic data. Once the decision is
made to develop a CVV, multiple strains may be available to serve as the basis for the CVV. In the event
that these strains have similar epidemiological and virological characteristics, the presence and type of
molecular markers for mammalian adaptation, transmissibility, and virulence can serve to differentiate
between strains. This application of GoF data enables more granular decision-making than would have
been possible based on other data sources alone, which is valuable because resource limitations constrain
that number of CVVs that can be produced.603
Both animal influenza viruses isolated from human infections as well as animal influenza viruses that
have not yet caused human infections can be subjected to a risk assessment (formally or informally).
However, because of the expense involved in each step of pre-pandemic vaccine production, none of the
above steps are likely to be undertaken unless multiple human infections have occurred.604 As a result,
although GoF approaches may aid the interpretation of surveillance data from animals, this proximal
benefit will not lead to downstream investments in pre-pandemic vaccine development but rather is
limited to deepening understanding of the risk associated with particular viruses.
A completely different strategy for increasing the availability of vaccines during a pandemic is by
shortening vaccine production timelines. GoF research that enhances virus production enables the
development of higher-yield CVVs, which shortens vaccine production timelines by increasing the rate of
bulk antigen production. Although this research can be immediately applied to improve vaccine
production, this strategy provides the greatest benefit to the production of vaccines using poor-growing
CVVs. However, as any strain may unexpectedly generate a low-yield CVV, such as the 2009 H1N1
pandemic strain, this benefit could significantly alleviate morbidity and mortality in the event that future
pandemic strains are also grow poorly.
601 Ibid.
602 (2015d) Influenza Vaccines. Interviews with Public Health Professionals Involved in Preventing and Responding to
604 (2015b) Interview with USG representative involved in pandemic risk assessment and decision-making about investments
pandemic preparedness initiatives.

Field investigations of clusters of zoonotic influenza infections abroad
The CDC participates in missions to investigate zoonotic influenza cases or clusters of concern abroad, in
conjunction with the WHO, OIE, Food and Agricultural Organization of the United Nations (FAO), and
local Ministries of Health. The goal of these missions is to supplement foundational surveillance with in-
depth investigations of ecological and environmental factors that may be contributing to spillover, such as
sources of human exposure and the extent to which the viruses are circulating in local animal
populations.605 Collectively, these data improve understanding of the risk posed by the zoonotic influenza
virus in that environment, which informs decision-making about other prevention and preparedness
activities (such as whether to develop a pre-pandemic CVV). Recent examples include missions to
Cambodia to investigate an abrupt rise in human H5N1 infections in 2013, to China in 2013 to investigate
the initial wave of H7N9 human infections, and to Cairo, Egypt in March of 2015 to investigate the
dramatic increase in the number of human cases of H5N1 infection recorded at the end of 2014 leading
into the first few months of 2015.606,607 The decision to send a CDC team abroad is informed by an
assessment of whether the sequences of human isolates contain molecular markers for mammalian
adaptation, virulence, and transmissibility. Similar to a formal risk assessment, this decision is driven by
epidemiologic data, but the presence of molecular markers of concern adds value by increases certainty in
decision-making. In addition, consideration of molecular marker data may stimulate increased attention to
investigations of the local animal population and human interactions with infected animals, undertaken to
better understand how ecological and environmental factors are influencing the evolution of the virus in
that area.
9.6.3.4 Vaccines
GoF approaches have the potential to benefit the development of pre-pandemic vaccines. Specifically,
pandemic risk assessments, which can be informed by GoF research (see Sec. 9.6.3.3), may trigger the
development of candidate vaccine viruses based on high-risk viruses, as well as subsequent stages of the
pre-pandemic vaccine production pipeline (e.g. manufacturing of clinical lot material, conducting human
clinical trials, and stockpiling vaccine).
9.6.3.5 Therapeutics
A lack of knowledge about whether existing therapeutics will be effective against future pandemic strains
hampers preparedness planning. GoF-generated viruses that are transmissible between ferrets may mimic
pandemic variants of that HA subtype better than wild type viruses. Thus, testing whether existing
therapeutics are capable of mitigating disease caused by GoF strains could inform pandemic preparedness
planning. Researchers have also suggested that these experiments could stimulate the development of new
therapeutics, in the event that existing therapeutics are found to be ineffective against GoF strains.
However, the relevance and utility of this information is severely constrained by several sources of
uncertainty, including a lack of knowledge about whether ferret-transmissible viruses are more
transmissible in humans, whether laboratory-generated transmissible viruses behave similarly to those
that could arise in nature, and other factors. Given this uncertainty, dedication of resources to developing
therapeutics targeting hypothetical future pandemic viruses is unlikely. Thus, this putative benefit to the
development of therapeutics is not considered in this report.
605 Davis CT et al (2014) Use of highly pathogenic avian influenza A(H5N1) gain-of-function studies for molecular-based
surveillance and pandemic preparedness. MBio 5
606 Ibid.

9.6.3.6 Diagnostics
Because the process of developing influenza diagnostics is well-established, GoF research does not
inform diagnostic development.608,609
Pandemic risk assessments inform prioritization of resources for pandemic preparedness. Specifically,
evaluating the relative risk posed by different influenza viruses helps decision-makers allocate limited
funds to pandemic preparedness efforts, such as the development of pre-pandemic vaccines targeting
high-risk viruses. This prioritization may improve the efficiency of government spending on influenza
pandemic preparedness. Economic benefits were not explicitly evaluated in this report.
9.6.4 Identification of the potential benefits and limitations of alt-GoF approaches that provide
similar potential benefits to the GoF approaches being examined
public health benefits as GoF approaches, but through different mechanisms. For each approach, the
scientific outcomes of the approach and how that information leads to similar benefits as GoF approaches
are described.
9.6.4.1 Potential Benefits and Limitations of alt-GoF to Scientific Knowledge
humans?
Characterizing the transmissibility of wild type isolates in representative animal models represents an
alternative approach for addressing whether animal influenza viruses display the capacity for transmission
between mammals. However, this approach is inherently reactive – that is, it can effectively answer
whether a virus is transmissible but cannot shed light on whether a virus has the potential to become
transmissible. Additionally, observations in animal models may not translate to humans.
Several alt-GoF approaches can address how influenza viruses evolve to efficiently infect and transmit in
humans. First, the comparison of sequences from closely related human and animal isolates enables the
identification of the origin and evolutionary rate of genetic changes among circulating viruses, which can
provide information on selection pressures and diversity among viruses in different hosts. That this
approach examines the natural course of adaptation and underlying mechanisms of infection and
transmission of viruses in humans is a strength relative to GoF approaches and other alternatives that
depend on the suitability of animal models in an artificial environment as representative of human

disease. However, this approach suffers from several significant limitations. The use of comparative
sequence analysis is feasible only if human-adapted and transmissible viruses have arisen in nature, but to
date, animal influenza viruses have limited capacity to infect and transmit in humans. Analysis of the few
animal-origin spillover infections may however inform evolution of adaptive traits. The success of this
approach is significantly constrained by the quality and availability of genetic surveillance data. In
particular, the noisiness of comparative sequence analysis due to high genetic diversity among influenza
viruses practically limits this approach to the examination of genetic regions known to be important for
adaptation and transmissibility, unless precursor-spillover paired strains can be identified (which is rare).
Additionally, the fact that the surveillance record is static and incomplete limits the depth of evolutionary
information that can be gleaned from this approach.
Analysis of viruses that have emerged from avian or mammalian reservoirs to become transmissible in
other mammalian species represents another surveillance-based approach for studying the mechanisms
underlying adaption to mammals during interspecies transmission. The recent emergence of animal
transmissible influenza viruses in other mammals (e.g. an avian-origin H3N2 canine influenza virus that
emerged in dogs in the mid-2000s) enables the study of the full evolutionary pathway for cross-species
acquisition of efficient transmissibility. This approach is subject to the same limitations as comparative
sequence analysis of human and animal isolates, with the additional caveat that adaptation to other
mammals may occur through different pathways and mechanisms than in humans.
Phenotypic characterization of wild type viruses by evaluating infectivity and transmissibility in

appropriate model systems is another alt-GoF approach for studying the evolution and mechanisms of
adaptation/transmissibility. This approach allows the generation of in-depth information about
evolutionary mechanisms; however, relevant evolutionary changes may not occur during a single round of
transmission. Moreover, any animal influenza viruses that are highly attenuated in representative animal
models or are incapable of establishing infection are not suitable for this approach. Finally, this approach
depends on the suitability of the animal models used for characterization.
9.6.4.1.3 Scientific Knowledge Gap #3: What are the genetic and phenotypic traits that result in adaption
and transmission in humans?
Several alt-GoF approaches can be used to uncover genetic and phenotypic traits underlying adaptation
and transmission in mammals. First, comparing the sequences of human and animal isolates enables the
identification of genetic changes that are associated with human adaptation and transmissibility. This
approach has the potential to directly identify human-adaptive traits and may be more likely to uncover
conserved traits through analysis of a large number of strains. However, the fact that no animal influenza
viruses that efficiently transmit in humans have been observed in nature precludes the use of this
approach to identify mechanisms underlying transmissibility. For the discovery of mammalian adaptive
traits, the success of this approach is constrained by the quality and availability of surveillance data. In
addition, the extensive genetic diversity within circulating virus populations and among viruses isolated
from humans makes discerning distinct genetic traits that are likely to contribute to fitness and
transmissibility in humans relative to animals difficult. Namely, the “noise” associated with sequences
comparisons obscures the discovery of relevant features that distinguish human versus animal isolates,
which practically limits this approach to the investigation of traits or regions previously known to be
important for adaptation.
Comparing the sequences of evolutionarily related isolates from different animal species represents
another surveillance-based approach for identifying genetic traits that are associated with mammalian
adaptation and transmissibility. Importantly, because avian-origin flu viruses that are airborne or contact
transmissible exist in circulation in several mammals including seals, horses, and dogs, this approach is
currently feasible for the study of transmissibility. In addition to the limitations above, mechanistic

insight gleaned through this approach may not translate to the adaptation of animal influenza viruses to
humans.
Phenotypic characterization of wild type viruses in appropriate animal models is another alt-GoF
approach that complements the use of surveillance data to study mechanisms underlying mammalian
adaptation and transmissibility. Specifically, comparing the sequences of wild type viruses with varied
levels of fitness and transmissibility enables the identification of genetic traits associated with
fitness/transmissibility. This approach is limited to the study of viruses that can productively infect and
transmit between animal models for adaptation/transmission. Notably, very few natural animal-origin
viruses are capable of transmission in ferrets and many are not able to efficiently cause disease in
representative animal models. Genetic and phenotypic traits uncovered through this approach may not
translate to human-adapted viruses and may only be applicable to the limited number of strains analyzed.
Loss-of-function (LoF) approaches, genetic screens that utilize random mutagenesis or targeted genetic
modification to identify genetic changes that attenuate fitness and transmission in mammals, can provide
information about genetic and phenotypic traits that contribute to transmissibility. Targeted LoF can also
be used to confirm necessary genetic or phenotypic traits by determining that mutations attenuate fitness
or transmission, but cannot identify traits that lead to enhanced transmission. This approach suffers from
several significant limitations. First, LoF studies can be performed only using transmissible seasonal or
pandemic viruses, and insights may not translate to animal influenza viruses. Second, because of the high
mutation rate of influenza viruses, LoF mutations that attenuate transmissibility may revert during the
single round of passage that is needed to characterize the transmissibility of the mutants (which represents
a selection step). Third, because many mutations attenuate transmission for trivial reasons, for example
mutations that compromise viability, discovering traits that directly contribute to transmissibility may be
difficult using a LoF approach. Finally, although in principle LoF screens can be performed after random
mutagenesis to discover new genetic elements important for transmission, the resource intensive nature of
transmission studies in ferrets practically limits these studies to the investigation of a few, known targets.
Several in vitro virus-free methods can be used to investigate phenotypes underlying adaptation and
transmissibility. Comparative sequence analysis of viral proteins with different phenotypic properties can
then enable the identification of mutations that are associated with relevant phenotypic changes, while
forward genetic screens can be used to identify novel genetic traits that contribute to underlying
phenotypes. Additional characterization involves the use of biochemical assays (e.g. characterizing the
acid stability of the HA protein) and crystallographic resolution of the structures of virus-host protein
complexes can provide insight into the functional and biophysical basis of underlying phenotypes. The
use of targeted modification of viral gene segments in isolation can also effectively confirm the necessary
and sufficient genetic traits that alter an underlying phenotype. Though the simplicity and relatively high-
throughput nature of these methods renders them appealing as a screening approach for the discovery and
confirmation of novel genetic traits that contribute to adaptation/transmissibility, these approaches are
inherently limited to the characterization of genetic traits and phenotypes previously identified in other
experiments. An additional drawback is that results gleaned from studying the behavior of a viral protein
or phenotype in isolation may not be recapitulated in the context of the full virus or in vivo. Moreover,
although fairly rapid phenotypic assays have been developed for the study of phenotypic traits known to
be associated with adaptation/transmissibility, assays to study phenotypic traits may be unreliable or
unavailable for future phenotypes of interest.
Structure-based modeling approaches, an in silico method, may also be used to predict the effects of
mutations on phenotypes underlying adaptation/transmissibility. This approach is critically limited by the
capabilities and accuracy of existing models, and as such any conclusions may not be consistent in the
context of the full virus.

Finally, several alt-GoF approaches focus on identifying host factors and host-virus interactions that are
associated with mammalian adaptation, which may provide indirect insight into viral mechanisms
underlying cross-species adaptation. Specifically, in vitro proteomic (e.g. mass spectrometry) and
genomic screens (e.g. RNAi screen) utilizing both virus-free and cell culture-based infection systems are
used to identify host factors that interact with virus proteins of interest or that are critical for underlying
phenotypes such as viral replication. These approaches complement the identification of viral
proteins/phenotypes underlying adaptation to new hosts. However, the breadth of proteomic approaches is
limited in that screens typically focus on a single viral protein, and both genomic and proteomic screens
can identify host proteins that may not be functionally relevant or may play minor roles in the viral life
cycle.
Another type of alternative approach involves the use of attenuated viruses for GoF methods, as a risk
mitigation strategy. Three types of attenuated viruses are used for such studies: reassortants with surface
protein gene segments from seasonal influenza viruses, to which the general population has pre-existing
immunity, reassortants with lab-adapted viruses (e.g. PR8), and strains which have virulence factors
altered or deleted (e.g. deletion of the multi-basic cleavage site in HPAI HA sequences). Results gleaned
through use of attenuated viruses may be of limited informational value because complex, multi-genic
traits depend on genetic context (a phenomenon called epistasis), and results may not be recapitulated in
the context of the wild type virus. Differences in disease pathogenesis, which critically influences the
biological processes of adaptation and transmission, further compromise the relevance of results gained
through the use of attenuated strains.
9.6.4.2 Benefits and Limitations of alt-GoF approaches to Surveillance
Akin to Sec. 9.6.3.2, this section evaluates the benefits of alt-GoF approaches for evaluation of the
infectivity, transmissibility, and virulence of animal influenza viruses detected through surveillance.
These virus properties may be directly measured in the laboratory or can be inferred from the viral genetic
sequence based on the presence of molecular markers that have been linked to those phenotypes through
previous research. Two other approaches are in development but are not yet used in public health practice
include the use of rapid assays to measure phenotypes underlying mammalian adaptation, transmissibility,
and virulence and computational modeling to predict phenotype from genotype. Each of these methods
has shortcomings which can be addressed by GoF approaches, as detailed in Sec. 9.6.3.2. The ability of
alt-GoF approaches to similarly address these shortcomings is evaluated below.
9.6.4.2.1 Analysis of alt-GoF approaches that support the development of rapid phenotypic assays
across many strain contexts. Moreover, interpretation of the results requires knowledge about how
these two questions constrain the development and use of rapid phenotypic assays. As discussed in detail
in Sec. 9.6.4.1.3 and Sec. 9.7.4.1.1, alt-GoF approaches can provide limited insight into these scientific
questions. The relevant findings are summarized below.
Characterization of wild type viruses provides limited insight into phenotypic traits underlying
mammalian adaptation and transmissibility because animal influenza viruses that efficiently infect and
transmit in humans do not exist in nature. However, characterizing the constellation of underlying
phenotypes present in a large number of wild type viruses (e.g. sialic acid receptor binding specificity,
HA stability, optimal temperature for polymerase activity, etc.) is uniquely capable of providing insight

into whether viruses that have a subset of the properties that are necessary for enhanced infectivity,
transmissibility, or virulence can persist in nature.
LoF approaches have limited utility for broad and unbiased identification of phenotypic traits that
contribute to transmissibility and pathogenicity due to their inefficiency and the fact that mechanisms
underlying transmissibility of seasonal/pandemic viruses may not translate to animal influenza viruses.
Though LoF approaches can be used to causally demonstrate that a particular phenotype is necessary for
efficient transmissibility and enhanced virulence, this approach cannot be used to understand how
multiple phenotypes synergize to enhance infectivity, transmissibility, or virulence. This information
critically informs how results from multiple phenotypic assays should be integrated to evaluate overall
pandemic potential. Surveillance-based approaches, including comparison of human and animal isolates,
comparison of sequences spanning avian to mammalian adaptation events, and comparison of viral
isolates with varying levels of virulence are limited to the study of previously known traits and provide
associative data. Notable exceptions include the analysis of precursor/spillover pairs, for the study of
adaptation/transmissibility, and analysis of viral isolates over the course of infection in a single patient,
for the study of virulence. However, the availability of both types of paired isolates is low. In addition,
surveillance-based approaches cannot provide insight into phenotypes underlying transmissibility because
animal influenza viruses that efficiently transmit in humans do not exist in nature. In vitro, virus free
approaches, which involve the study of known phenotypes in isolation, cannot provide information about
the functional relationships among underlying phenotypes or between underlying phenotypes and
adaptation/transmissibility.
9.6.4.2.2 Analysis of alt-GoF approaches that support the use of molecular markers to evaluate the risk
posed by circulating animal influenza viruses
As described previously, the current utility of molecular markers for the interpretation of genetic
surveillance data is constrained by multiple sources of scientific uncertainty. As discussed in detail in Sec.
9.6.4.1.1 and Sec. 9.6.4.1.3, alt-GoF approaches can provide some insight into relevant scientific
questions that strengthen this approach. The relevant findings are summarized below.
In sum, alt-GoF approaches, namely characterization of wild type viruses, are uniquely capable of
demonstrating whether partially adapted viruses exist in nature, which provides insight into whether
complex phenotypes such as adaptation, transmissibility, and virulence can accrue in a step-wise fashion
(an underlying assumption of the use of molecular markers to evaluate pandemic risk). However, alt-GoF
approaches have significant limitations for addressing other relevant knowledge gaps at the phenotypic
level, in particular strengthening the linkage between underlying phenotypes and mammalian
adaptation/transmissibility/virulence.
Alt-GoF approaches can provide some insight into the scientific knowledge gaps about the genetic traits
underlying mammalian adaptation, transmissibility, and virulence that compromise the application of
molecular marker data to surveillance. Characterization of wild type viruses provides limited insight into
genetic traits underlying mammalian adaptation and transmissibility because animal influenza viruses that
efficiently infect and transmit in humans do not exist in nature. LoF approaches have limited utility for
broad and unbiased identification of novel genetic traits that are necessary for transmissibility or
enhanced virulence due to their inefficiency and the fact that mechanisms underlying transmissibility of
seasonal/pandemic viruses may not translate to animal influenza viruses. Surveillance-based approaches,
including comparison of human and animal isolates and of sequences spanning avian to mammalian
adaptation events, have limited utility for the discovery of novel genetic traits associated with
adaptation/transmissibility/virulence due to the high genetic diversity of influenza viruses and
shortcomings in the quality and availability of surveillance data. However, surveillance-based approaches
have several unique strengths for validating the functional consequences of particular markers.

Comparison of human and animal isolates or of human isolates with varying levels of virulence is
uniquely capable of providing direct insight into traits associated with human adaptation and virulence
across multiple strain contexts. These traits can be considered “causally” linked if a large enough number
of sequences are compared. Notably, this approach cannot be used to validate markers associated with
enhanced transmissibility because animal influenza strains that transmit efficiently between humans do
not exist in nature. The high-throughput nature of in vitro, virus free approaches relative to animal
experiments renders them appealing for the discovery of additional mutations that give rise to particular
phenotypic changes (through forward genetic screens) and for validating the function of particular
markers in new genetic contexts. However, results may not be recapitulated in vivo, in the context of the
full virus.
9.6.4.2.3 Analysis of alt-GoF approaches that improve predictive models
Another strategy for evaluating the infectivity, transmissibility, or virulence of involves the use of
computational models that predict phenotypes underlying mammalian adaptation, transmissibility, and
virulence based on viral sequences. However, existing computational models cannot reliably predict
underlying phenotypes based on sequence information.
about mutations that do and do not give rise to phenotypic changes of interest. This data is critical for
building models that can account for the context dependence of genetic changes in influenza biology.
Alternative experimental approaches cannot provide this information. In addition, model predictions must
be validated experimentally, which feeds back to improve model accuracy. Alternative approaches can
only test model predictions using in vitro, virus-free systems. As results may not be recapitulated in the
context of the full virus, this approach is of limited utility for improving the quality of models.
Experimental data about the biophysical basis of underlying phenotypes, such as crystallography data and
measurements of HA binding affinities to α2,6 and α2,3 sialoglycans, is also needed to improve existing
models. This information can only be generated using alt-GoF approaches.
9.6.4.3 Benefits and Limitations of GoF approaches to Inform Policy Decisions
GoF approaches have potential to benefit decision-making in public health policy by contributing to
pandemic risk assessments, which guide investments in pandemic preparedness initiatives, as described in
Sec. 9.6.3.3.2. This section evaluates how alternative types of data contribute to pandemic risk
assessments, thereby similarly benefitting downstream decision-making related to pandemic
preparedness. Three alternative data sources are described: virologic data, epidemiologic data, and
ecological data.
9.6.4.3.1 Potential benefits and limitations of alt-GoF approaches to pandemic risk assessments
Potential Benefits and Limitations of Alternative Pandemic Risk Assessment Factors: Virologic data
The relative strengths and weaknesses of using virological approaches to characterize the phenotypic
properties of surveillance viruses were discussed extensively in Sec. 9.6.3.2.1. This section evaluates the
utility and limitations of virologic data in the context of the overall pandemic risk assessment.
Several risk elements rely on laboratory data: receptor binding (preference for “human-like” α2,6
sialylated receptors, “avian-like” α2,3 sialylated receptors, or dual specificity), transmission in animal
models, antiviral resistance, disease severity in animal models, and antigenic relationship between virus

and existing CVVs/vaccines. 610,611 Although epidemiologic measurements also provide information about
the severity and transmissibility of a virus, these phenotypes are difficult to measure accurately in nature,
especially when a virus first emerges in human populations and epidemiological data are scarce. As
performing human transmission and virulence studies using novel influenza viruses would be unethical,
laboratory-generated phenotypic data critically complement epidemiologic observations. Accordingly, in
a recent assessment of three influenza viruses (an avian H1N1 virus, a human isolate of H7N9, and a
human isolate of H3N2v), virologic data contributed highly to the overall risk score. For evaluating the
likelihood of emergence, laboratory data about transmission and sialic acid receptor binding specificity
were about two-thirds and half as important as the extent of human infections, respectively. For
evaluating potential consequences of emergence, disease severity, which reflects the severity of human
infections as well as the severity in appropriate animal models, was most important.612 The major
limitation associated with laboratory-generated phenotypic data is that political, logistical, and regulatory
factors delay receipt of clinical specimens/viral isolates in U.S. labs and subsequent generation of
virologic data.
Epidemiologic data
Two risk elements rely on epidemiologic data: human infections, disease severity (which is also informed
by laboratory testing in animals), and population immunity (detection of pre-existing cross-reactive serum
antibodies). The human infections element considers the number and frequency of human cases and
evidence for human-to-human transmission, while the disease severity element considers the spectrum of
illness observed in humans, including the age distribution of deaths. The human infections and disease
severity elements are the most important elements of the likelihood and consequences components of the
IRAT, respectively, because the data directly reflect the properties of the virus in humans. However, there
are several challenges associated with the interpretation of epidemiological data for pandemic risk
assessments. When a novel virus first emerges, extrapolating virus properties from a limited number of
human cases may be difficult. In particular, disease severity is often initially over-estimated because only
severe cases interact with the public health system, and serological studies to ascertain population
exposure are difficult and time-consuming to carry out.
Ecological/environmental factors
Finally, two risk elements involve ecological factors, which collectively consider the global distribution
of the virus in animals, the number of species that can be and are infected, and the potential extent of
exposure between humans and those animal species. Other environmental information, such as the
strength of the public health systems and the strength of the relationship between the public health and
veterinary services sectors in countries in which the virus is circulating in animal populations, may also
be considered. These elements moderately contribute to the likelihood component and minimally
contribute to the consequence component of the IRAT. Importantly, these elements reflect completely
different aspects of risk than the elements based on phenotypic, genetic, and epidemiologic data.
9.6.4.3.2 Public health impacts of pandemic risk assessments: development of pre-pandemic vaccines
The development of pre-pandemic vaccines will lead to earlier vaccine availability during a pandemic,
thereby reducing human morbidity and mortality, as discussed in Sec. 9.6.1.3.3.3. Several completely
different strategies can be used to increase the availability of vaccines during a pandemic, thus achieving

the same ultimate public health goal. These strategies are described in detail in Sec. 9.5.4.2.2 and are
briefly summarized here. First, a universal or broad-spectrum flu vaccine could be deployed in advance of
a pandemic or could be rapidly deployed following the emergence of a novel pandemic strain. However,
influenza and vaccinology experts disagree about the scientific feasibility of developing a universal
vaccine, and one expert felt that a ten to twenty year time frame for development is optimistic. Second,
several scientific and technical advancements could shorten production timelines for strain-specific
vaccines, which would lead to faster vaccine availability during a pandemic. New vaccine platforms, such
as recombinant vaccines, can be rapidly scaled up and have shorter production timelines than egg- and
cell-based vaccines. However, the one recombinant vaccine on the market accounts for less than 1% of
total seasonal influenza vaccine produced annually, and although several other virus-free vaccine
platforms are in development, the length and expense of licensure processes for new vaccines will delay
their widespread availability. Incorporating adjuvants into existing egg- and cell-based vaccines would
allow for a smaller quantity of antigen to be used per vaccine dose, thus enabling production of the same
number of doses in a shorter period of time. However, only one US-licensed pandemic vaccine includes
adjuvants. Although an active area of research, adjuvanted vaccines must undergo standard FDA
licensing procedures for new vaccines and thus are unlikely to be broadly available in the near future.
approaches
transmissibility relative to alt-GoF approaches are discussed, in each benefit category that GoF
approaches can address.
humans?
GoF approaches are uniquely capable of proactively assessing the potential for any animal influenza
viruses to acquire enhanced fitness and transmissibility in mammals. Notably, the relevance of this
information for human populations depends on the suitability of animal models as well as whether
laboratory-acquired mutations can arise in nature, both of which are unknown.
GoF approaches are uniquely capable of providing in-depth information about the evolution of
mammalian fitness/transmissibility in any animal influenza virus strain. In addition, GoF approaches are
uniquely capable of demonstrating the order(s) of acquisition of genetic changes that are necessary and
sufficient to lead to enhanced fitness/transmissibility in mammals. However, the relevance of information
derived from GoF approaches is contingent upon how well animal models represent human disease and
how well the lab environment mimics natural evolution.
For those wild type strains that are naturally capable of productively infecting laboratory animals used for
transmission studies, simply characterizing the transmissibility of a strain in animals, an alt-GoF
approach, has the potential to generate similarly in-depth information. However, a single round of
transmission may be insufficient for relevant adaptive changes to accrue or may reveal only part of the
adaptive process, which further lessens the relative utility of this alt-GoF approach. Surveillance-based
approaches, including comparison of human and animal isolates and comparison of animal isolates from
different species, are uniquely capable of reporting on the real-world evolution of a variety of strains, thus

complementing two shortcomings of GoF approaches. Though results gleaned from comparative analysis
of human and animal isolates are directly translatable to humans, the fact that animal influenza virus
strains that efficiently transmit in humans have not been observed in nature precludes use of this approach
for the study of transmissibility in particular. While case studies of interspecies transmission events exist,
the translatability of that information to the evolution of human adaptive traits is uncertain.
9.6.5.1.3 Scientific Knowledge Gap #3: What are the genetic and phenotypic traits that result in adaption
GoF approaches are uniquely capable of identifying novel genetic and phenotypic traits underlying
mammalian adaptation and transmissibility in any animal influenza virus strain of interest. Furthermore,
targeted genetic modification of viruses to introduce genetic traits associated with mammalian
adaptation/transmissibility is uniquely capable of demonstrating that particular genetic markers are
necessary and sufficient for mammalian adaptation and transmissibility across multiple virus contexts.
Given the importance of genetic context for influenza biology, this approach critically strengthens the
certainty of scientific knowledge about mechanisms underlying mammalian adaptation and
transmissibility. However, results gleaned from cell culture and animal model studies may not translate to
human disease. Notably, attenuated strains cannot be used to study mechanisms underlying airborne
transmission because these strains do not efficiently infect ferrets.
Characterizing wild type viruses, an alt-GoF approach, also has the potential to uncover previously
unknown traits. However, the fact that this approach cannot be used to study animal influenza viruses that
do not productively infect laboratory animals and that relevant changes may not arise during a single
round of transmission renders it less useful than GoF approaches. LoF approaches have limited utility for
broad and unbiased identification of necessary genetic and phenotypic traits due to their inefficiency and
the fact that mechanisms underlying transmissibility of seasonal/pandemic viruses may not translate to
animal influenza viruses. The simplicity and relative high-throughput nature of in vitro, virus-free
systems renders them appealing for the discovery of novel genetic traits that alter known phenotypes
underlying mammalian adaptation/transmissibility, but properties observed may not be recapitulated
during the complete viral life cycle.
Unlike GoF methods, the use of human and animal surveillance data for the discovery of genetic markers
associated with adaptation/transmission directly translates to human disease and has strength in numbers
as it analyzes genetic traits across large data sets. Critically, this approach cannot be used for studying
transmissibility because animal or zoonotic viruses that efficiently transmit in humans have not been
observed in nature. Analysis of sequences spanning avian to mammalian adaptation events enables the
identification of “real-world” markers associated with mammalian adaptation/transmissibility but may not
translate to human-adapted viruses. For both surveillance-based approaches, shortcomings in the quality
and availability of surveillance data compromise the feasibility of this approach and the relevance of any
findings.
Finally, host-focused approaches, such as proteomic and genomic screens, cannot supplant the
identification of viral adaptation/transmissibility traits but rather complement GoF approaches by
identifying host factors that contribute to those processes.
9.6.5.2 Benefits to Surveillance
A key goal of influenza surveillance is to monitor the evolution of circulating animal influenza viruses, in
order to identify those viruses that pose a risk of emerging in human populations to cause a pandemic.
Resources can then be dedicated to mitigating the risks that those viruses emerge and the potential
consequences of an emergence event. Analysis of the phenotypic properties of individual surveillance

isolates is an important aspect of pandemic risk assessments, including transmissibility and virulence in
mammals. Currently, this analysis relies on the laboratory characterization of surveillance isolates and, to
a lesser extent, the inspection of sequences for molecular markers associated with phenotypes underlying
mammalian adaptation, transmissibility and virulence. Both methods exhibit shortcomings that
compromise the accuracy, timeliness, and quantity of data. Two additional approaches have been
proposed: the development of rapid assays for phenotypes underlying mammalian adaptation and
transmissibility, and the use of computational models to predict underlying phenotypes from genotype.
Such rapid phenotypic assays do not yet exist, and the prospective accuracy of existing models is
unknown. Both GoF and alt-GoF experimental approaches have potential to address shortcomings
associated with the use of rapid phenotypic assays, molecular markers, and computational models,
thereby benefitting surveillance of animal influenza viruses.
GoF approaches provide unique benefits to the design and validation of rapid assays for phenotypes
underlying adaptation, transmissibility, and virulence. The fact that these assays would be high-
throughput and less technically challenging than ferret experiments could increase the quantity and
timeliness of phenotypic data available, relative to the use of traditional phenotypic characterization
assays for adaptation, transmissibility, and virulence. The accuracy and utility of rapid phenotypic assays
depends on establishing a strong linkage between underlying phenotypes and
adaptation/transmissibility/virulence as well as developing an understanding of how multiple phenotypes
synergize to enhance the infectivity, transmissibility, and virulence of animal influenza viruses in
mammals. GoF approaches represent the most efficient and effective approach for discovering novel
phenotypes underlying mammalian adaptation, transmissibility, and virulence and are uniquely capable of
demonstrating that a particular phenotype is causally linked to enhanced
infectivity/transmissibility/virulence in mammals across multiple virus contexts. GoF approaches are also
uniquely capable of causally determining how multiple underlying phenotypes interact to enhance
infectivity, transmissibility, or virulence in mammals, which provides insight into how information about
underlying phenotypes should be integrated for a risk assessment. However, a major caveat associated
with GoF approaches is that results gleaned from laboratory studies involving animal models may not
translate to human disease in nature. Characterizing the constellation of underlying phenotypes present in
a large number of wild type viruses (alt-GoF) is uniquely capable of providing insight into whether
partially adapted viruses can persist in nature, which lends support to the practice of inferring complex
phenotypes such as adaptation, transmissibility, and virulence based on data about underlying phenotypes.
In addition to the need for scientific advancements, a notable barrier to realization of the benefits derived
from the use of rapid phenotypic assays is that these assays must be carried out under BSL-3 conditions,
which limits the number of diagnostic laboratories that will be able to conduct the assays.
GoF approaches provide unique benefits to the practice of using molecular markers to infer phenotypes
underlying adaptation/transmissibility/virulence based on genetic sequence data. The fact that sequence
data can be reliably generated at NICs and other diagnostic labs in developing countries can increase the
timeliness and quantity of phenotypic data available, relative to the conduct of traditional phenotypic
characterization assays at WHOCCs. Currently, most molecular markers for mammalian adaptation,
transmissibility, and virulence have low predictive value due to significant scientific uncertainties
associated with the association between underlying phenotypes and adaptation/transmissibility/virulence,
whether the function of markers is conserved across different strain contexts, and incomplete knowledge
about the breadth of mutations that can give rise to a particular phenotypic change. As discussed above,
GoF approaches provide essential data for strengthening the linkage between underlying phenotypes and
adaptation/transmissibility/virulence. GoF approaches also provide unique advantages for discovering
novel markers and strengthening the predictive value of known markers. Namely, GoF approaches
represent the most efficient and effective approach for discovering novel genetic traits underlying
mammalian adaptation, transmissibility, and virulence and are uniquely capable of demonstrating that
particular genetic markers are necessary and sufficient for mammalian adaptation, transmissibility, or

virulence across multiple virus contexts. However, the validation of molecular markers for mammalian
adaptation or virulence through analysis of genetic surveillance data (alt-GoF) is uniquely capable of
providing direct insight into traits associated with human adaptation/virulence across multiple strain
contexts, which complements GoF approaches. Notably, surveillance-based approaches are not viable for
the validation of molecular markers associated with transmissibility because animal influenza strains that
transmit efficiently between humans in nature do not exist. GoF approaches are also uniquely capable of
systematically exploring alternative mutational pathways for altering an underlying phenotype in the
context of whole virus. In vitro, virus free approaches can also be used, but results may not be
recapitulated in the context of the full virus. As above, the major caveat associated with GoF approaches
is that results gleaned from laboratory studies involving animal models may not translate to human
disease in nature. Finally, in addition to the need for scientific advancements, a notable barrier to the full
realization of benefits derived from the use of molecular markers is the need to expand sequencing
capabilities at NICs.
GoF approaches are also critical for improving models for prediction of underlying phenotypes based on
sequence data. Specifically, GoF approaches that generate information about mutations that do and do not
give rise to phenotypic changes of interest provide critical training data for models, and GoF approaches
are needed to validate model predictions in the context of the full virus. Importantly, other types of
biophysical data generated through alternative experimental approaches are also critical for improving the
accuracy of existing models. Similar to the use of molecular markers, full realization of the benefits
derived from the use of computational models will require significant scientific advancements as well as
the expansion of sequencing capabilities at NICs.
Both the direct measurement of virus phenotypes in the laboratory and the prediction of underlying
phenotypes from genotype, either through sequence inspection for molecular markers or computational
modeling approaches, have inherent strengths and limitations. Namely, the generation of phenotypic data
will always be delayed by the need to ship virus samples, but direct measurements of phenotypic
properties are invaluable. In contrast, as sequence data is increasingly available from NICs and other
“base” level diagnostic laboratories, the application of predictive methods will enable the rapid generation
of phenotypic “data” that reflects the properties of viruses present in clinical samples, allowing for more
rapid characterization of emerging influenza viruses. However, due to the inherent uncertainties
associated with predictions, the subsequent confirmation of predictions through phenotypic testing is
critical. Therefore, virological data and sequence-based predictive data are complementary, and
consideration of both will strengthen the timeliness and accuracy of assessments of virus properties that
contribute to pandemic risk.
9.6.5.3 Benefits to pandemic risk assessment, decision-making in public health policy
GoF approaches have potential to benefit pandemic risk assessments by strengthening the predictive value
of molecular markers for mammalian adaptation, transmissibility, and virulence, which are a component
of the “genomic variation” risk element considered in the assessment. The importance of this element
relative to other risk elements places a qualitative “upper bound” on the potential benefits of GoF research
to pandemic risk assessments. Notably, because molecular marker data are currently incorporated into
pandemic risk assessments, the benefits of GoF-derived improvements to the reliability of molecular
marker data could be immediate.
Epidemiological data (alt-GoF) represent the most important input to the risk assessment, for both the
likelihood and consequences of emergence component of the IRAT. Laboratory data about
transmissibility and virulence in appropriate animal models and receptor binding specificity also
significantly contribute to the overall pandemic risk score. Genomic variation, which includes
consideration of molecular marker data for mammalian adaptation, transmissibility, and virulence, is

relatively less important. Given the caveats associated with epidemiological and virological data, subject
matter experts involved in the pandemic risk assessment process emphasized the value of corroborating
information about infectivity, transmissibility, and disease severity in humans or appropriate animal
models with molecular marker data.613 That genetic data can increase confidence in an estimate of risk
adds certainty to decision-making downstream of the risk assessment, which is valuable.
Molecular marker data play a more important role in the risk assessment when a novel influenza virus
first emerges in the human population. In this scenario, epidemiological data will be scant and sequence
data are likely to be available before phenotypic data. As a result, the use of molecular marker data
enables a rapid risk assessment of the emerging virus, so that downstream response actions can be
initiated more quickly if deemed appropriate. In the event of a pandemic, such a three to four week head
start on vaccine production could significantly reduce pandemic-associated morbidity and mortality. For
example, researchers estimate that deployment of vaccine two weeks earlier during the 2009 H1N1
pandemic would have prevented an additional ~600,000 cases (an approximately 60% increase in the
number of cases prevented), while deployment of the vaccine four weeks earlier would have prevented an
additional 1.4 million cases (an approximately 135% increase in the number of cases prevented).614
Once the decision is made to develop a CVV, multiple strains may be available to serve as the basis for
the CVV. In the event that these strains have similar epidemiological and virological characteristics, the
presence and type of molecular markers for mammalian adaptation, transmissibility, and virulence can
serve to differentiate between strains. This application of GoF data enables more granular decision-
making than would have been possible based on other data sources alone, which is valuable because
resource limitations constrain that number of CVVs that can be produced.
International surveillance for influenza is improving, especially in the wake of the 2009 pandemic, but
gaps remain, particularly in certain regions of the world (e.g. parts of Africa, regions experiencing
political instability, etc.). The limited breadth of available surveillance data constrains the potential
benefits of using pandemic risk assessments to guide decision-making about pandemic preparedness
investments. That is, experts can evaluate and prepare for pandemics caused by strains they know about
only. For that reason, all stakeholders interviewed for this report, including influenza researchers, public
health personnel, and USG public health policy representatives, agreed that there is a clear need to
strengthen and expand influenza surveillance networks. Importantly, expanded surveillance alone is not
sufficient to improve pandemic risk assessments without concomitant improvements to the tools used for
pandemic risk assessments, including the use of molecular marker data. Thus, strong surveillance
networks function as a co-factor that is needed for the full realization of GoF benefits to pandemic risk
assessments.
As discussed in Sec. 9.6.3.2, GoF approaches can also benefit surveillance for animal influenza viruses by
enabling the development of rapid assays for phenotypes underlying mammalian adaptation,
transmissibility, and virulence, as well as by improving computational models for sequence-based
predictions of underlying phenotypes. Either type of data could be used to corroborate information about
transmissibility and virulence gleaned through ferret experiments. Given the variability inherent in animal
experiments, data about underlying phenotypes could strengthen the robustness of this phenotypic
information. However, the timeline for realization of this benefit is likely to be long-term. The benefits
arising from rapid phenotypic assays depends on the discovery and validation of suitable underlying
phenotypes and the development and validation of an appropriate rapid phenotypic assay. The benefits
arising from the use of computational models depend on the development of reliable models, which will

likely prove to be a significant scientific challenge. The timescales for these scientific and technical
innovations are unknown.
9.7 Influenza viruses: Benefits of GoF research that enhances virulence
9.7.1 Summary
This section describes the benefits of GoF research that is reasonably anticipated to enhance the virulence
of influenza viruses in representative animal models. Such GoF studies were found to generate scientific
knowledge; to inform surveillance of circulating animal influenza viruses, which has downstream impacts
on decision-making about USG investments in pandemic preparedness initiatives; and to inform the
development of new influenza vaccines and therapeutics. Alt-GoF approaches that may generate similar
benefits were also identified and analyzed. At present, GoF studies resulting in enhanced virulence in
mammals have unique benefits to scientific knowledge, surveillance, and pandemic preparedness, though
full realization of GoF benefits to public health requires significant scientific advancements. Chapter 9.7
provides an overview of these benefits, including basic background and supporting information; a fully
referenced and more thorough discussion of these benefits can be found in Appendix IV Sec. 15.4.
9.7.1.1 Benefits of GoF that enhances pathogenicity to scientific knowledge
• GoF approaches:
o Are the most efficient and effective strategies for identifying novel viral genetic and
phenotypic traits underlying pathogenicity and are uniquely capable of demonstrating that a
particular viral trait is necessary and sufficient to enhance virulence. However, results in
model systems may not translate to human disease.
o Are capable of identifying host factors that are associated with enhanced pathogenicity.
o Are capable of generating animal models that recapitulate human disease for the study of
pathogenicity through adaptation of viruses to host animals. However, adaptive mutations
may alter the biology of the virus.
o Are uniquely capable of providing direct insight into genetic traits associated with enhanced
virulence in humans, but are severely constrained by the quality and availability of existing
surveillance data.
o Are capable of demonstrating that particular viral traits are necessary for virulence, which
complements GoF approaches.
o Are uniquely capable of confirming that a particular host factor contributes to virulence
and/or deleterious host immune responses.
o Are capable of generating animal models for the study of pathogenicity and to support MCM
development through sensitization of host animals to viral infection through targeted gene
knockout or the use of immunosuppressants. However, results using immunocompromised
animals may not translate to healthy hosts.

9.7.1.2 Benefits of GoF that enhances pathogenicity to surveillance
• GoF approaches:
o Provide a foundation for the development of rapid assays for phenotypes underlying
virulence, which have potential to increase the quantity and timeliness of phenotypic
information about animal flu viruses detected through surveillance. However, the success of
this approach is subject to significant advancements in the state of knowledge about
mechanisms underlying pathogenicity.
o Are uniquely capable of strengthening the predictive value of molecular markers for
virulence, which can be used to infer phenotype from sequence. Use of molecular markers in
lieu of or to corroborate phenotypic testing results could improve the quality, timeliness, and
quantity of phenotypic information about animal flu viruses detected through surveillance.
knowledge the mechanistic basis of pathogenicity, and predictions must be experimentally
validated.
o Are critical for improving computational models for predicting phenotypes underlying
pathogenicity based on sequence, which could improve the quantity and timeliness of
phenotypic information about animal flu viruses detected through surveillance. However, the
success of this approach is subject to significant advancements in the state of knowledge the
mechanistic basis of pathogenicity, and predictions must be experimentally validated.
o Have significant limitations for advancing the development of rapid assays for phenotypes
underlying pathogenicity and for strengthening the predictive value of molecular markers for
virulence.
o Are also critical for improving computational models for predicting phenotypes underlying
pathogenicity based on sequence, but through the generation of different types of data that
complement data generated through GoF approaches.
o Phenotypic assays for virulence are uniquely capable of providing direct information about
this complex phenotype under controlled conditions, but results may be delayed relative to
the publication of viral sequences or, in the future, the generation of data about underlying
phenotypes through rapid assays.
9.7.1.3 Benefits of GoF that enhances pathogenicity to decision-making in public health policy
• GoF approaches:
o Molecular markers for virulence, which can be generated and validated using GoF
approaches, moderately influence pandemic risk assessments of circulating animal influenza
viruses, relative to other types of data that are considered. Pandemic risk assessments guide
downstream decisions about investments in pre-pandemic vaccines, which will increase
vaccine availability during a pandemic if a similar strain emerges to cause a pandemic.
o Molecular marker data plays a relatively important role when novel influenza viruses first
emerge in human populations, when epidemiological data are scarce and virological data are
not yet available. The ability to conduct a rapid risk assessment using molecular marker data
can provide a three to four week head start on vaccine production.

o Molecular marker data can guide selection of particular viruses to use as the basis of pre-
pandemic vaccines, when multiple viruses have similar epidemiological and virologic
characteristics.
o Epidemiological data are the most influential data in a pandemic risk assessment, but disease
severity can difficult to accurately measure in human populations, and epidemiological data
may be scarce when novel viruses first emerge in human populations.
o Virologic data strongly influence pandemic risk assessments, but the generation of virological
data may be delayed relative to the publication of sequencing data when novel viruses emerge
abroad, due to shipping delays.
o Other types of data, such as ecological data, also contribute to pandemic risk assessments but
completely molecular marker data (GoF) by evaluating completely different aspects of
pandemic potential.
9.7.1.4 Benefits of GoF that enhances pathogenicity to vaccine development
• GoF approaches:
o Are uniquely capable of determining whether live attenuated influenza vaccine (LAIV)
candidates recover virulence upon growth in cells or animals, an important aspect of safety
testing. LAIVs are being explored as potential pandemic vaccines for avian influenza viruses
and have shown promise.
o Can be used to discover genetic traits that confer enhanced virulence, which can be removed
from vaccine viruses to increase the safety of vaccine production. However, alt-GoF
approaches must first be used to demonstrate that mutating particular virulence markers is
sufficient to attenuate the virulence of vaccine viruses.
o Are capable of generating animal models that recapitulate human disease to support vaccine
development through adaptation of viruses to host animals. However, adaptive mutations may
alter the susceptibility of the virus to vaccines, rendering results mis-representative.
o Several alternative vaccine platforms are also being explored as potential pandemic vaccines
for avian influenza viruses and have shown promise.
o Are uniquely capable of determining that mutating or deleting particular virulence markers
attenuates the virulence of vaccine viruses, which can improve the safety of vaccine
production.
o Are capable of generating animal models for the study of pathogenicity and to support
vaccine development through sensitization of host animals to viral infection through targeted
gene knockout or the use of immunosuppressants. However, results using
immunocompromised animals may not translate to healthy hosts.

9.7.1.5 Benefits of GoF that enhances pathogenicity to the development of therapeutics
• GoF approaches:
o Represent the most efficient and effective strategies for identifying novel viral factors that
contribute to virulence, which may be good targets for new therapeutics.
o Are capable of generating animal models that recapitulate human disease to support
therapeutic development through adaptation of viruses to host animals. However, adaptive
mutations may alter the susceptibility of the virus to therapeutics, rendering results mis-
representative.
o Represent the most effective strategies for identifying novel host factors that contribute to
virulence, which may be good targets for new therapeutics.
o Other alternative approaches for the development of new therapeutic candidates, including
high-throughput screening of small molecule compounds and selection of monoclonal
antibodies that bind to particular virus proteins, are also being actively pursued and have
generated promising therapeutic candidates.
o Are capable of generating animal models for the study of pathogenicity and to support
vaccine development through sensitization of host animals to viral infection through targeted
gene knockout or the use of immunosuppressants. However, results using
immunocompromised animals may not translate to healthy hosts.
9.7.2 Overview of GoF research landscape: enhanced pathogenicity in representative animal

models
9.7.2.1 Serial passaging of viruses in cell culture or animal models
Serial passaging of viruses in cell culture or animals selects for viruses with enhanced fitness or virulence,
respectively. This approach is performed for two purposes: (1) to develop animal models for studying the
mechanistic basis of flu-associated morbidity/mortality and for medical countermeasure development, and
(2) to identify mutations that are associated with enhanced fitness/virulence, which provides a foundation
for follow-up studies that investigate the mechanistic basis of pathogenicity. These studies can also
provide insight into host mechanisms underlying disease pathology by correlating host immune responses
with morbidity and mortality measures. Both types of serial passaging studies are performed with
seasonal and animal (i.e. avian and swine) viruses, and animals such as mice, ferrets, and swine may be
used.
9.7.2.2 Forward genetic screen to identify mutations that enhance fitness/virulence
fitness/virulence or comprehensive reassortment of parental gene segments from two viruses, followed by
characterization of the fitness or virulence of mutants in appropriate mammalian model systems to select
for mutant viruses with enhanced fitness/virulence. Sequencing emergent viruses enables the
identification of mutations or gene segments that enhance the fitness/virulence of viruses, which provides
a foundation for follow-up studies that investigate the mechanistic basis of pathogenicity in mammals.
These studies are performed using human seasonal viruses, the 1918 H1N1 pandemic virus, and animal
viruses. A variant of this approach involves the use of strains with impaired fitness due to the evolution of

antiviral resistance, to determine whether strains can recover fitness through the acquisition of
compensatory mutations, which has been performed using seasonal strains.
9.7.2.3 Targeted modification of viruses to introduce traits that are expected to enhance
fitness/virulence in mammals

introduce genetic traits that are expected to enhance the fitness/virulence of viruses followed by
characterization of the fitness/virulence of mutants in cell culture or animal model systems, respectively,
may lead to the generation of viruses with enhanced fitness/virulence in mammals. This approach is
phenotypic trait, such as evasion of a particular innate immune response, contributes to the complex
phenotype of pathogenicity, and (2) to confirm that a particular mutation or gene segment is necessary
and sufficient to enhance the fitness/virulence of viruses in appropriate model systems. Traits that are
associated with enhanced pathogenicity may be discovered through GoF approaches, such as serial
passaging, or alt-GoF approaches, such as random mutagenesis followed by screening for attenuated
virulence (loss-of-function). As above, this information provides a foundation for follow-up studies
investigating the mechanistic basis of pathogenicity. These studies are performed using human seasonal
viruses, the 1918 H1N1 pandemic virus, and animal viruses.
Here we evaluate the potential benefits of GoF research that enhances fitness and pathogenicity in each
benefit category listed in the NSABB Framework are discussed
GoF approaches have potential to benefit scientific knowledge in several ways. First, GoF approaches can
provide insight into the mechanistic basis of pathogenicity, including the identification of viral and host
traits that contribute to pathogenicity. Second, GoF approaches enable the identification of compensatory
mutations that rescue the growth of antiviral resistant strains, which provides a foundation for follow up
studies investigating the mechanistic basis of the enhanced fitness phenotype. Finally, viruses with
enhanced virulence developed using GoF approaches can be used as tools to understand how the host
immune response contributes to morbidity and mortality observed during influenza infections. The
benefits and limitations of GoF approaches in each of these scientific areas is addressed in more detail
below.

9.7.3.1.1 Scientific Knowledge Gap #1: What are the viral genetic and phenotypic traits that underlie
pathogenicity in mammals? What are the host factors that contribute to enhanced pathogenicity,
as well as infection-associated morbidity and mortality?
Introduction
The pathogenesis of influenza viruses reflects the complex interactions between viral and host factors and
is the result of both the virus’s ability to cause disease and the host’s response to viral infection. While
advances in research have revealed functions of specific influenza proteins and genetic traits that
contribute to virulence, considerable gaps in knowledge remain about the molecular basis and the role of
each underlying phenotype in in defining pathogenicity and associated disease outcomes. Moreover, there
is limited understanding of the host factors that contribute to protective versus deleterious outcomes.
Insight into virus-host interactions is needed to advance in-depth understanding of virulence and
pathogenesis of influenza viruses.
Benefits and limitations of GoF approaches
Several GoF approaches can be used to discover the genetic and phenotypic markers underlying enhanced
pathogenicity of influenza viruses:
to pathogenicity by either site-directed mutagenesis or targeted reassortment (often between
animal-origin or human pandemic and human seasonal strains),

by selection for enhanced virulence, or underlying phenotypes, and
• Serial passaging in appropriate animal models or mammalian cells to select for viruses with
enhanced pathogenicity.
enhanced pathogenicity in representative model systems. The GoF approaches described here also
provide insight into host response pathways that contribute to underlying disease pathology. Serial
passaging has the potential to uncover novel viral genetic and phenotypic traits that contribute to
enhanced virulence. In contrast, because forward genetic screens involving random mutagenesis typically
focus on regions that are suspected or known to play a role in phenotypes underlying pathogenicity, this
approach can discover novel viral genetic markers for enhanced virulence only. The targeted genetic
modification approach is limited to the investigation of viral genetic traits and underlying phenotypes that
are suspected to contribute to pathogenicity (e.g. determining whether enhanced polymerase activity
contributes to pathogenicity).
Targeted genetic modification is also used to confirm that particular virus mutations or gene segments are
necessary and sufficient to enhance virulence in mammals. Often this experiment is followed by
characterization of other virus phenotypes, such as infectivity and tissue tropism. Furthermore, this
approach provides associative insight into how host responses are altered during infection with the
modified strain. Collectively, this information provides a strong foundation for follow-up studies
investigating the mechanistic basis of pathogenicity, including the study of host-virus interactions.
Taken together, these GoF studies provide a foundation for follow-up cell biological, immununological,
and pathological studies that elucidate the mechanistic basis of viral factors contributing to virulence,
corresponding host responses, and how both factors alter susceptibility to secondary bacterial infection.

Additionally, GoF approaches permit the identification of host immune responses that are associated with
enhanced pathogenicity. Although the analysis of host factors contributing to enhanced pathogenicity is
indirect, this information can be derived from the comparison of genetically similar virus backgrounds
displaying a dynamic range of virulence (i.e. GoF and parental strains). The relevance of these approaches
depends on whether mechanisms underlying enhanced virulence in cell culture and animal models are
representative of those in humans. Another drawback of these approaches is that results gleaned from the
study of one or a few strains may not be recapitulated in different genetic contexts.
9.7.3.1.2 Scientific knowledge gap #2: Provide insight into whether fitness defects associated with the
acquisition of antiviral resistance can be overcome, and the mechanisms underlying recovery of
fitness
Introduction
Though influenza viruses can readily mutate to acquire resistance to therapeutics, antiviral-resistant
viruses are often initially less fit than parental viruses. The relative fitness of antiviral-resistant strains has
implications for how likely and how quickly these strains are to spread in nature. Whether and how
antiviral strains can acquire compensatory mutations that enhance fitness while preserving the antiviral
resistance phenotype is unknown for most antiviral resistance mutations. Studies investigating this
question provide insight into the mechanistic basis of viral fitness and the mechanistic interplay between
antiviral resistance and other virus phenotypes.
Benefits and limitations of GoF approaches
Several GoF approaches can be used to determine whether antiviral-resistance strains with impaired
growth can recover fitness and to identify compensatory mutations that rescue growth, which provides a
foundation for follow-up biochemical and cell biological studies that investigate the mechanistic basis of
enhanced growth. First, growth-impaired strains can be serially passaged in cells or animals to select for
strains with enhanced fitness, following by sequencing of emergent viruses to identify genetic changes
that arose. However, this approach often results in reversion of antiviral-resistance mutations rather than
the evolution of compensatory mutations. A second approach involves forward genetic screens to identify
mutations that are sufficient to rescue fitness. While this approach is more likely to uncover compensatory
mutations than serial passaging, screening large libraries of mutants is relatively labor-intensive,
particularly if mutations are introduced into multiple virus proteins (as compensatory mutations may arise
in proteins that do not contain antiviral-resistance mutations). Finally, targeted mutagenesis is used to
confirm that a particular mutation or set of mutations is necessary and sufficient to rescue the fitness of a
growth-impaired strain.
9.7.3.1.3 Scientific Knowledge Benefit #3: Generation of animal models for the study of flu-associated
morbidity/mortality and for vaccine and therapeutic development
Model systems that can be efficiently infected by influenza viruses and exhibit the spectrum of disease
observed during human infections are essential for the study of influenza-associated morbidity/mortality.
Serial passaging of influenza viruses in laboratory animals, which enhances the virulence of the virus in
that animal, generates animal models that can be used to study the mechanisms underlying the
pathogenesis of influenza viruses. This approach is performed for two purposes: (1) to generate viruses
capable of efficiently infecting mice, as mice are inherently resistant to infection with human seasonal
influenza viruses and some animal influenza viruses, and (2) to generate viruses with enhanced
pathogenicity, if wild type viruses exhibit a limited spectrum of disease in representative animal models.
In particular, the generation of mouse models is useful for pathogenesis studies due to the wide array of
immunological and other experimental tools that have been developed for mice. However, the passaging

needed to adapt the virus to representative animal models may alter the biology of the virus, such that
results do not translate to natural disease.
9.7.3.2 Benefits and Limitations of GoF approaches to Surveillance
GoF approaches that lead to the identification of molecular markers for enhanced pathogenicity have the
potential to inform the interpretation of wildlife, agricultural animal, and public health surveillance
information. Specifically, determining the presence (or absence) of particular mutations associated with
enhanced virulence is one aspect of evaluating the risk posed by circulating animal influenza viruses, as
viral virulence plays a key role in the expected public health consequences caused by a novel influenza
virus emerging in human populations. The strategies for monitoring the virulence of animal influenza
viruses detected through surveillance are similar to those for monitoring mammalian adaptation and
transmissibility, and GoF approaches that enhance virulence and those that enhance infectivity and
transmissibility in representative animal models benefit surveillance through similar mechanisms. Thus,
these benefits are discussed collectively in Sec. 9.6.3.2.
GoF approaches that lead to the identification of compensatory mutations that rescue the fitness of
antiviral-resistant strains with impaired growth do not benefit surveillance. Because of the high mutation
rate of influenza viruses, influenza surveillance experts expect that antiviral resistant strains that initially
exhibit impaired fitness can readily acquire compensatory mutations that rescue growth. Thus, experts
simply track the presence of antiviral resistance markers, and the additional presence of absence of a
known compensatory mutation does not increase or decrease the level of risk associated with the antiviral
resistance marker.
9.7.3.3 Benefits and Limitations of GoF to the Development of Vaccines
GoF approaches have potential to benefit the development of vaccines in three ways:
• Serial passaging of candidate live attenuated vaccine strains in animals is used to test whether
strains recover virulence upon growth in vivo, which is an important aspect of vaccine safety.
• GoF approaches enable the identification of conserved virulence determinants in the HA and NA
proteins. These markers may be removed vaccine viruses through targeted deletion or
mutagenesis, as is commonly done for the multi-basic cleavage site present in the HA proteins
from some avian influenza strains, which may improve the efficacy and safety of the vaccine
production process.
• Animal models developed using GoF approaches can be used for testing the safety and efficacy
of vaccine candidates.
9.7.3.3.1 Vaccine development benefit 1: Development of new influenza vaccine candidates
Introduction – current strategies and challenges for developing pandemic vaccines for avian influenza
viruses
Standard methods for production of seasonal influenza vaccines have posed challenges for the production
of vaccines targeting highly pathogenic avian influenza strains such as H5N1.615 In addition, egg-based
production systems are not amenable to rapid scale-up due to their reliance on the egg supply, which
615 Baz M et al (2013) H5N1 vaccines in humans. Virus Res 178: 78-98

would pose a major problem if a novel pandemic virus emerged off production cycle. For these reasons,
researchers are exploring a variety of other platforms for the production of vaccines for avian influenza
viruses with pandemic potential. Live attenuated influenza vaccines (LAIVs) are attractive for pandemic
vaccines for several reasons related to their efficacy and relative ease of production and administration. 616
However, a major concern associated with LAIVs is their potential to regain virulence in people, through
reversion or the acquisition of compensatory mutations.617
Potential benefits and limitations of GoF approaches: LAIV safety
Because of the concern that LAIVs could regain virulence in people, the WHO recommends serial
passaging of LAIV candidates during the non-clinical phase of in vivo toxicity and safety testing (a GoF
approach), to determine whether the LAIV is genetically stable or recovers virulence upon passage in
animals.618, 619 In accordance with these recommendations, multiple candidate LAIVs have been subjected
to serial passaging in animals.620,621,622, 623
9.7.3.3.2 Vaccine development benefit 2: targeted mutagenesis to remove virulence markers from vaccine
viruses
Background – challenges for production of vaccines for highly pathogenic avian influenza viruses
Removal of the multi-based cleavage site from the HA protein of highly pathogenic avian influenza
(HPAI) strains, a major determinant of viral virulence, is standard practice for the production of HPAI
vaccines.624 This mutagenesis further attenuates the vaccine virus (which is also attenuated through
reassortment with an attenuated vaccine backbone strain such as PR8), enabling safe and efficient
production of vaccine in eggs (or cells) under BSL-2 conditions. In the future, other conserved
determinants of virulence in the HA and NA proteins of avian influenza (AI) viruses could be similarly
deleted from AI vaccine viruses in order to further improve the safety of the vaccine production process.
Potential benefits and limitations of GoF approaches
As discussed above (Sec. 9.7.3.1.1), GoF approaches, in particular forward genetic screens and serial
passaging, represent efficient and effective methods for discovering novel viral genetic and phenotypic
traits that contribute to virulence. This information provides a foundation for follow-up LoF studies that
aim to determine how to attenuate virulence, the goal of vaccine virus development, through mutation or
deletion of those traits.
616 Ibid.
617 Ibid.
618 WHO Expert Committee on Biological Standardization. (2010) Recommendations to assure the quality, safety and efficacy
of influenza vaccines (human, live attenuated) for intranasal administration. WHO Technical Report Series No 977, 2013.
The World Health Organization,, Geneva, Switzerland pp. 163-196.
619 The World Health Organization. (2005) WHO guidelines on nonclinical evaluation of vaccines. WHO Technical Report
Series, No 927, 2005, Geneva, Switzerland, pp. 32-63.
620 Jang YH, Seong BL (2012) Principles underlying rational design of live attenuated influenza vaccines. Clinical and
experimental vaccine research 1: 35-49
621 Han P-F et al (2015) H5N1 influenza A virus with K193E and G225E double mutations in haemagglutinin is attenuated and
immunogenic in mice. Journal of General Virology 96: 2522-2530
623 Sedova ES et al (2012) Recombinant influenza vaccines. Acta Naturae 4: 17-27

9.7.3.4 Benefits and Limitations of GoF to the Development of Therapeutics
GoF approaches have potential to benefit the development of influenza therapeutics in two ways:
• GoF approaches that provide insight into viral and host traits that contribute to virulence identify
potential targets for next-generation therapeutics (either targeting the virus or the host), and
• Animal models developed using GoF approaches can be used for testing the safety and efficacy
of therapeutic candidates.
9.7.3.4.1 Therapeutic development benefit: inform the development of next-gen therapeutics
Only one class of licensed antivirals is recommended for use in the U.S., the neuraminidase inhibitors
(NAIs).625 Mutations that confer resistance to one or multiple NAIs have been observed in nature, though
are not yet widespread, and the NAIs exhibit limited efficacy.626 Thus, there is an urgent need for the
development of new therapeutics against influenza viruses.627 Researchers are actively working to
develop next-gen influenza therapeutics that directly target viral proteins as well as therapeutics that
inhibit host factors that are critical for viral virulence or that exacerbate infection-associated pathology.
GoF approaches have potential to benefit the development of both types of therapeutics.
9.7.3.4.2 Potential benefits of GoF to therapeutic development
As discussed in detail in Sec. 9.7.3.1.1, GoF approaches represent the most efficient and effective
strategies for discovering novel viral genetic traits that contribute to pathogenicity, which may be good
targets for novel therapeutics. In addition, targeted genetic modification of viruses to introduce traits
associated with pathogenicity is uniquely capable of demonstrating that particular viral genetic traits are
necessary and sufficient to enhance virulence across multiple virus contexts, which provides a strong
mechanistic basis for the role of that viral factor in virulence.
GoF approaches also enable the identification of host factors that are associated with virulence and
immunopathology, which may be good targets for novel host-targeted therapeutics. However, because the
GoF approach is indirect, the role of a particular host protein in virulence/immunopathology must be
confirmed using alt-GoF approaches, which provides an important conceptual foundation for the design
of therapeutics targeting that protein. Nonetheless, targeted modification to introduce mutations that are
expected to enhance pathogenicity (GoF) provides a controlled system for studying the interplay between
virus and host factors that contribute to pathogenicity, which is a valuable complement to alt-GoF
approaches that perturb the function of host factors, a more blunt approach. Notably, in both cases,
whether inhibiting viral or host factors discovered through GoF approaches is sufficient to attenuate viral
replication or infection-associated pathology must be empirically determined using alt-GoF approaches.
9.7.3.5 Benefits and limitations of GoF to both vaccine and therapeutic development: enable the
development of MCMs
9.7.3.5.1 Background – shortcomings in existing influenza vaccines and therapeutics
Shortcomings in existing influenza vaccines and therapeutics compromise public health preparedness for
influenza pandemics and exacerbate the public health consequences of annual influenza epidemics,
625 CDC. Influenza Antiviral Medications: Summary for Clinicians. http://www.cdc.gov/flu/professionals/antivirals/summary-

clinicians.htm. Last Update November 3, 2015. Accessed November 28, 2015.
626 Ibid.
627 (2015l) Interviews with influenza researchers.

highlighting the need for development of new influenza vaccines and therapeutics. Testing the safety and
efficacy of candidate MCMs in animal models is a critical aspect of MCM development. Mice, a common
animal model used for the development of influenza MCMs, are naturally resistant to infection with many
influenza viruses. GoF or alt-GoF approaches can be used to develop animal models to study the
effectiveness of MCMs against these viruses. The development of MCMs that protect against severe
disease necessitates testing the efficacy of candidate MCMs in animal models that exhibit exacerbated
disease pathology. In cases where wild type viruses cause a limited spectrum of disease, GoF or alt-GoF
approaches may be used to generate model systems that display a larger dynamic range of virulence.
9.7.3.5.2 Potential benefits and limitations of GoF approaches: determining MCM safety and efficacy
GoF approaches to generate new model systems for characterizing the safety and efficacy of MCMs
involve serial passaging of viruses in animals to enhance the infectivity and virulence of the virus toward
that host. Two variants of this approach support MCM development. First, as mice are naturally resistant
to many influenza viruses, passaging of those viruses in mice generates a model system for testing the
efficacy of MCMs against that virus. Second, passaging of virus in ferrets to enhance virulence generates
a model system exhibiting exacerbated pathology, which can be used to screen MCM candidates for their
ability to protect against severe disease.628 One key strength of this approach is that comparing the
efficacy of MCMs following challenge with two genetically similar viruses provides certainty that
differences in outcomes reflect true distinctions between the function of MCMs rather than disparate
interactions with genetically different viruses. The main drawback associated with these approaches is
that the changes that accrue during passaging may alter the susceptibility of the virus to the MCM under
study, thus compromising the relevance of any results.
9.7.3.6 Benefits and Limitations of GoF to Diagnostics
inform diagnostic development.629
9.7.3.7 Benefits and Limitations of GoF to Inform Policy Decisions
GoF approaches that lead to the identification of molecular markers for enhanced pathogenicity contribute
to assessments of the pandemic risk posed by circulating animal influenza viruses, which are based on
genetic surveillance data and several other types of data (e.g. epidemiologic data, phenotypic data, etc.).
These assessments inform policy decisions related to public health preparedness for novel influenza
outbreaks, including whether to develop pre-pandemic vaccines. This GoF benefit to decision-making in
public health policy is discussed in detail in Sec. 9.6.3.3.2, as evaluation of the transmissibility of animal
influenza viruses similarly informs pandemic risk assessments and downstream decision-making.
GoF benefits to the development of new vaccines and therapeutics could have downstream economic
benefits. We did not explicitly evaluate economic benefits in this report.
628 Ibid.

public health benefits as GoF approaches, but through different mechanisms. For each approach, the
scientific outcomes of the approach and how that information leads to similar benefits as GoF approaches
are described.
9.7.4.1 Potential Benefits and Limitations of alt-GoF to Scientific Knowledge
Several alt-GoF approaches can be used to uncover genetic and phenotypic traits underlying
pathogenicity in mammals. First, comparing the sequences of human isolates that display varying degrees
of pathogenicity enables the identification of genetic changes that are associated with increased virulence.
Unlike the GoF approaches described above, this approach has the potential to directly identify genetic
traits that contribute to pathogenicity in humans and may be more likely to uncover conserved traits
through analysis of a large number of strains. However, this approach is subject to significant limitations
relative to GoF approaches. First, the utility of this approach is significantly constrained by the quality
and availability of existing surveillance data. Second, the use of consensus sequences in standard
surveillance practices may not be able to uncover genetic traits that are present at low frequencies in
human populations. Finally, the extensive genetic diversity within circulating virus populations makes
discerning distinct viral genetic traits that are likely to contribute to pathogenicity difficult, which
practically limits this approach to the investigation of traits or regions previously known to be important
for pathogenicity.
A variant of the surveillance-based approach involves corroboration of sequence data with

immunopathological observations from autopsies, which provides an opportunity to identify host factors
or genetic polymorphisms that are broadly associated with severe disease.630 In addition to the limitations
described above, this approach is limited by the availability of autopsy data and is subject to the caveat
that autopsies represent late stage, lethal disease, which may not be representative.
Comparing the sequences of isolates within patients, over the course of infection and/or from different
tissue sources, represents another approach for identifying genetic traits that contribute to pathogenicity in
humans. Specifically, comparing early and late isolates during prolonged disease and comparing isolates
from the primary site of infection (i.e. the upper respiratory tract) and those from disseminated sites (i.e.
lower respiratory tract), which are associated with increased virulence, enables the identification of
adaptive mutations that enhance virulence. A strength of this approach is that the reduced viral genetic
diversity observed within a single patient may enable the identification of novel genetic traits associated
with virulence. However, this approach is limited to the analysis of viral isolates from patients presenting
with severe disease, which may not be representative.
Phenotypic characterization of wild type viruses in appropriate cell culture or animal models is another
alt-GoF approach that can be used to study mechanisms underlying pathogenicity in mammals.
630 Everitt AR et al (2012) IFITM3 restricts the morbidity and mortality associated with influenza. Nature 484: 519-523

Specifically, comparing the sequences of wild type viruses with varied levels of fitness in vitro and
pathogenicity in vivo enables the identification of genetic and phenotypic traits associated with increased
virulence. Similar to GoF approaches, this approach can also identify host response pathways that are
associated with varying disease outcomes, including susceptibility to secondary infection. Because of the
high genetic diversity among existing viral isolates phenotypic characterization is often limited to the
analysis of known determinants of pathogenicity unless highly genetically similar strains are available.
Additionally, the use of in vivo models is restricted to the study of viruses that can productively infect
representative animal model systems, which excludes some animal-origin viruses with low fitness. (Such
strains are typically passaged in mice for adaptation prior to analysis of virulence, which represents a GoF
approach.) As for the GoF approaches, genetic and phenotypic traits uncovered through this approach
may not translate to humans.
modification to identify changes that attenuate fitness/virulence, can also provide information about
genetic and phenotypic traits that contribute to pathogenicity. The screening approach has the potential to
identify novel genetic traits associated with pathogenicity, while the targeted approach is used to confirm
whether particular genetic traits are necessary for pathogenicity. This information complements that
generated by GoF methods, but LoF approaches suffer from several limitations. First, because of the high
mutation rate of influenza viruses, LoF mutations that attenuate pathogenicity may revert during the
single round of passage that is needed to characterize the virulence of the mutants (which represents a
selection step). Second, although in principle, LoF screens for mutations that attenuate virulence can be
performed in an unbiased manner, characterizing the pathogenicity of a large panel of mutants in animals
is labor-intensive and expensive. As a result, the use of this method may be practically limited to cell
culture systems or the investigation viral phenotypes previously shown to be associated with
pathogenicity. Third, because many mutations attenuate pathogenicity for trivial reasons, for example
mutations that compromise viability, discovering traits that directly contribute to virulence in high
pathogenicity strains relative to low pathogenicity strains may be difficult using a LoF approach.
The use of replication incompetent viruses provides another alternative method for the identification of
genetic and phenotypic traits underlying pathogenicity.631 In these model systems, viral replication and
immune evasion pathways, both of which contribute to pathogenicity in vivo, can be assessed in cell
culture lines that are engineered to stably express an essential viral protein that is missing from the
“replication-incompetent” virus strains used for infection. The result is a virus that is biologically
constrained to replication in that cell line, which therefore poses low risk to people.632,633 Using these
systems, viruses can be serially passaged to identify novel adaptive mutations that are associated with
phenotypes underlying pathogenicity. However, cell culture systems cannot provide information about the
effect of identified genetic traits on global host responses, virus dissemination, and associated morbidity
and mortality. Additionally, in vitro results may not be recapitulated during in vivo infection.
Several in vitro virus-free methods can be used to investigate phenotypes underlying pathogenicity. Cell
biological assays (e.g. measuring viral polymerase activity) and crystallographic resolution of the
structures of viral protein interactions with other viral or host factors (e.g. virus-host protein-protein
complexes) can provide insight into the mechanistic and biophysical basis of underlying phenotypes.
Comparative sequence analysis of viral proteins with different phenotypic properties can then enable the
631 The use of this approach has been proposed during interviews with influenza researchers as a possible method, although the
use of this approach for explicitly identifying genetic and phenotypic viral and host factors contributing to fitness and cell-
specific immune evasion is currently limited.
632 Martínez-Sobrido L et al (2010) Hemagglutinin-Pseudotyped Green Fluorescent Protein-Expressing Influenza Viruses for
the Detection of Influenza Virus Neutralizing Antibodies. Journal of virology 84: 2157-2163
633 Rimmelzwaan GF et al (2011) Use of GFP-expressing influenza viruses for the detection of influenza virus A/H5N1
neutralizing antibodies. Vaccine 29: 3424-3430

identification of mutations that are associated with relevant phenotypic changes or provide insight into the
molecular basis for virus-host interactions. Alternatively, forward genetic screens can be used to identify
novel genetic traits that contribute to underlying phenotypes, while targeted modification of viral gene
segments in isolation confirms the set of genetic changes that are necessary and sufficient to alter an
underlying phenotype. Though the simplicity and relatively high-throughput nature of these methods
renders them appealing as a screening approach for the discovery of novel genetic traits associated with
pathogenicity, these approaches are inherently limited to the investigation of previously identified viral
phenotypes. An additional drawback is that results gleaned from studying the behavior of a viral protein
or phenotype in isolation may not be recapitulated in the context of the full virus or in vivo. Although
fairly rapid phenotypic assays have been developed for the study of phenotypic traits known to be
associated with pathogenicity, assays to study certain phenotypic traits may be unreliable or unavailable
for future phenotypes of interest.
The use of in silico approaches to model the biophysical properties of viral proteins, virus-host and virus-
virus protein complexes can be used to evaluate mutations that may alter phenotypes underlying
pathogenicity. Although this approach may provide insight into the biophysical basis of interactions
underlying phenotypes of interest, the success of the approach is limited by the accuracy of existing
models.
Finally, because pathogenicity reflects virus-host interactions, several alt-GoF approaches focus on
identifying and characterizing host factors that are associated with pathogenicity, which may provide
indirect insight into viral mechanisms underlying virulence in representative animal models. The use of
transcriptional (e.g. qRT-PCR, microarray) and translational (e.g. ELISA) expression profiling, as well as
immunophenotyping (e.g. identifying the type and kinetics of immune cell recruitment) and
histopathology, independently or in the context of the GoF and alt-GoF approaches discussed above, can
identify host response pathways that change during infection and thus may play a role in pathogenicity.
Another host-focused approach involves in vitro proteomic (e.g. mass spectrometry) and genomic screens
(e.g. RNAi screen) utilizing both virus-free and cell culture-based infection systems to discover host
factors that interact with virus proteins of interest or that are critical for underlying phenotypes such as
viral replication and immune evasion. These approaches provide direct insight into host factors involved
in viral fitness. However, screens may identify host proteins that are not be functionally relevant or may
play minor roles in the viral life cycle in vivo. Following the discovery of host factors or signaling
pathways that may play a role in pathogenesis, genetically modified mouse lines (e.g. knockout mice) or
pharmacological inhibitors can be used to confirm the role of a particular protein, signaling pathway, or
immune cell type in pathogenicity. Taken together, the strength of these approaches is that they provide
direct information about host factors involved in pathogenicity. However, given the complexity of the
immune response to influenza virus infection, resolving the function of particular host proteins in the
context of globally altered host factors and regulatory networks may be difficult.
A second type of alternative approach involves the use of attenuated viruses, as a risk mitigation strategy.
Three types of attenuated viruses are used for such studies: (1) reassortants with surface protein gene
segments from seasonal influenza viruses, to which the general population has pre-existing immunity, (2)
reassortants with lab-adapted viruses (e.g. PR8), and (3) strains which have virulence factors altered or
deleted (e.g. deletion of the multi-basic cleavage site in HPAI HA sequences). The use of reassortants
with lab-adapted strains to identify viral determinants that are necessary and sufficient to enhance
virulence in a low-pathogenicity background is possible, as many of these strains are well characterized
and provide a large dynamic range for evaluating increases in virulence. Despite those advantages, the
results gleaned through use of attenuated viruses are subject to the caveat of epistasis. That is, because
complex, multi-genic traits depend on genetic context, causative genetic and phenotypic traits that
contribute to enhanced virulence in attenuated strains may not be recapitulated in the context of other wild
type strains and interactions with other factors (not present in the attenuated strain) may contribute to

virulence. Similarly, differences in disease pathogenesis relative to wild type viruses further compromise
the relevance of results gained through the use of some attenuated strains, in particular if the mechanism
of attenuation alters phenotypes underlying virulence.
9.7.4.1.2 Scientific knowledge gap #2: Discover whether fitness defects associated with the acquisition of
antiviral resistance can be overcome, and the mechanisms underlying recovery of fitness
Two alt-GoF approaches can be used to identify compensatory mutations that may rescue the growth of
antiviral-resistant strains with impaired fitness. First, comparative analysis of the sequences of antiviral-
resistant strains with varying levels of fitness may enable the identification of mutations that are
associated with enhanced fitness. However, due to the high genetic diversity among influenza viruses,
generating strong hypotheses about mutations that are linked to the recovery of fitness is difficult. In
addition, this approach is reactive, limited to the discovery of compensatory mutations after antiviral-
resistant strains have recovered growth in nature. A second approach involves computational modeling to
predict mutations that may rescue the fitness of growth-impaired strains. However, all predictions must be
experimentally confirmed using targeted mutagenesis, a GoF approach. Additionally, because existing
computational models cannot predict epistasis effects, the in silico approach is limited to the discovery of
compensatory mutations that arise in the same protein carrying the antiviral-resistance mutations.
9.7.4.1.3 Scientific benefit #3: Generation of animal models for the study of flu-associated
morbidity/mortality and for vaccine and therapeutic development
Alt-GoF approaches to develop animal models for the study of influenza pathogenesis involve increasing
host susceptibility to infection through the use of inbred mouse lines, knockout/transgenic mice, or the
treatment of mice with immunosuppressants. This approach can enable the study of wild type viruses that
do not efficiently infect wild type mice. The use of genetic modification is largely limited to the use of the
mouse model system, for which there are a broad array of well-established tools. However, the mouse
model is less representative of human disease than other animal models, such as the ferret. The use of
immunosuppressants is a promising alternative. The key limitation of this approach is that results gleaned
through the use of immunocompromised hosts may not translate to healthy human populations.
9.7.4.2 Benefits and Limitations of alt-GoF approaches to Surveillance
Circulating animal influenza viruses detected through surveillance of humans and animals are monitored
for their potential infectivity, transmissibility, and virulence in human populations, as these properties
inform the likelihood that viruses will evolve to efficiently infect and transmit in humans and the expected
public health consequences of their emergence in human populations. The strategies for monitoring the
virulence of viruses detected through surveillance are similar to those for monitoring mammalian
adaptation and transmissibility, and include strategies that are informed by GoF approaches and those that
are independent of GoF. The latter set of approaches is discussed in Sec. 9.6.4.2.
9.7.4.3 Benefits and Limitations of alt-GoF to the Development of Vaccines
9.7.4.3.1 Vaccine development benefit 1: Development of new influenza vaccine candidates
Due to shortcomings in existing methods for influenza vaccine production, researchers are exploring a
variety of other platforms for the production of vaccines for avian influenza viruses with pandemic
potential.634 While GoF approaches inform the development of LAIVs, several alternative vaccine
platforms which do not rely on GoF for their development, such as recombinant vaccines, are also being

explored. These vaccine platforms have strengths and limitations relative to LAIVs. For example,
adjuvanted, inactivated vaccines may provide broad-spectrum immunity but require multiple doses to
confer high levels of protection.635
9.7.4.3.2 Vaccine development benefit 2: Establish LAIV safety
GoF approaches are used to demonstrate that candidate LAIVs do not recover virulence upon growth in
cells or animals, an important aspect of safety testing. There are no alternative approaches that can
provide similar information.
9.7.4.3.3 Vaccine development benefit 3: targeted mutagenesis to remove virulence markers from vaccine
viruses
The HA multibasic cleavage site is removed from vaccine viruses based on HPAI strains to enable their
propagation in eggs and to improve the safety of the vaccine production process. Deletion of other
conserved determinants of virulence in the HA and NA proteins of avian influenza (AI) viruses could
further improve the safety of the vaccine production process in the future.
Several alt-GoF approaches can be used to discover novel virulence factors, including comparative
analysis of surveillance data, comparative analysis of the sequences of wild type viruses with varying
levels of virulence, use of replication incompetent viruses, and LoF forward genetic screens. As discussed
above, each of these approaches has critical limitations for the discovery of novel virulence traits relative
to GoF approaches. However, following the identification of novel genetic traits that contribute to
virulence, targeted mutagenesis can be used to identify particular mutations within that genetic region that
lead to attenuated virulence in multiple virus strains, which is essential for application of the information
to the vaccine development process.
9.7.4.4 Benefits and Limitations of alt-GoF to the Development of Therapeutics
GoF approaches have potential to inform the development of new candidate therapeutics for influenza
viruses. Several alt-GoF approaches, described below, also have potential to inform the development of
new influenza therapeutics.
As discussed in detail in Sec. 9.7.4.1.1, alt-GoF approaches have significant limitations for the discovery
of novel viral genetic traits and factors that contribute to virulence. However, alt-GoF approaches play a
critical role in establishing the function of putative virulence traits, which complements mechanistic
information that can be gleaned through GoF approaches. In particular, targeted LoF can be used to
confirm that blocking or attenuating the function of a particular virulence factor is sufficient to attenuate
viral replication and/or infection-associated pathology. This information establishes an evidence base for
efforts to design therapeutics targeting that virulence factor.
Alt-GoF approaches provide valuable insight into host factors that enhance pathogenicity and contribute
to deleterious immune responses. Specifically, the use of targeted knockout animals or pharmacological
inhibition of the host factor during infection is uniquely capable of confirming that a host factor
contributes to virulence and pathogenicity. Other alt-GoF approaches may be used to gain further
mechanistic insight into the role of the host factor during infection, including characterization of host
immune responses to identify host genes that are up-regulated during infection and LoF targeted genetic
modification of viruses to tease apart the role of particular virus-host interactions in pathogenesis.636
635 Ibid.
636 Cheung CY et al (2002) Induction of proinflammatory cytokines in human macrophages by influenza A (H5N1) viruses: a
mechanism for the unusual severity of human disease? Lancet 360: 1831-1837

Taken together, these studies provide a conceptual foundation for the design of therapeutics targeting that
protein.
In addition to designing therapeutics targeting specific virulence factors or pathways (virus or host),
several alternative strategies are used to develop novel candidate therapeutics. One alternative approach
for designing new therapeutics involves high-throughput screening of small molecule compounds to
identify compounds that reduce viral replication in vitro, which may identify candidate therapeutics that
target viral or host proteins.637,638 This approach has generated promising candidates, including
therapeutics that are in Phase III clinical trials in the U.S.639 One drawback of this approach is that it is
limited to the identification of compounds that reduce viral replication, which is only one aspect of
virulence.
Another alternative approach involves identifying neutralizing monoclonal antibodies (mAbs) targeting
virus proteins. These approaches isolating mAbs that bind to particular virus proteins, such as the HA
protein, the nucleoprotein (NP), the NA protein, and the M2 protein, from the B cells of convalescent
patients or of mice that have been injected with the virus protein of interest.640,641,642,643,644 Subsequently,
the ability of mAbs to neutralize virus activity is tested. This approach has also generated promising
therapeutic candidates, including therapeutics that have entered Phase I clinical trials.645,646 However,
mAb-based therapeutics have several drawbacks, including high production costs and the need for
injection-based delivery.647
9.7.4.5 Benefits and limitations of alt-GoF to both vaccine and therapeutic development: enable the
development of MCMs
GoF approaches have potential to benefit the development of new influenza vaccine and therapeutics by
enabling the development of animal models that can be used to test the safety and efficacy of MCM
candidates. Alt-GoF approaches, described below, can also be used to development animal models that
support MCM development.
Alternative approaches for the development of new model systems involving sensitizing the host to
infection through targeted genetic modification (use of inbred mouse lines or knockout/transgenic mice)
637 Furuta Y et al (2002) In vitro and in vivo activities of anti-influenza virus compound T-705. Antimicrobial agents and
chemotherapy 46: 977-981
638 An L et al (2014) Screening and identification of inhibitors against influenza A virus from a US drug collection of 1280
drugs. Antiviral research 109: 54-63
639 Toyama Chemical Company, Ltd. Pipeline. https://www.toyama-chemical.co.jp/en/rd/pipeline/index.html. Last Update
640 Krause JC et al (2011a) A broadly neutralizing human monoclonal antibody that recognizes a conserved, novel epitope on
the globular head of the influenza H1N1 virus hemagglutinin. Journal of virology 85: 10905-10908
641 Clementi N et al (2011) A human monoclonal antibody with neutralizing activity against highly divergent influenza
subtypes. PloS one 6: e28001
642 Bodewes R et al (2013) In vitro assessment of the immunological significance of a human monoclonal antibody directed to
the influenza a virus nucleoprotein. Clinical and vaccine immunology : CVI 20: 1333-1337
643 Shoji Y et al (2011) An influenza N1 neuraminidase-specific monoclonal antibody with broad neuraminidase inhibition
activity against H5N1 HPAI viruses. Human vaccines 7 Suppl: 199-204
644 Grandea AG, 3rd et al (2010) Human antibodies reveal a protective epitope that is highly conserved among human and
nonhuman influenza A viruses. Proceedings of the National Academy of Sciences of the United States of America 107:
12658-12663
645 HHS funds 2 experimental flu treatments. CIDRAP. http://www.cidrap.umn.edu/news-perspective/2015/09/hhs-funds-2-
experimental-flu-treatments. Last Update September 29, 2015. Accessed November 8, 2015.
646 Visterra Pipeline. http://www.visterrainc.com/pipeline/pipeline.html. Last Update Accessed November 8, 2015.

or the use of immunosuppressants (in ferrets or mice).648,649,650,651 A strength of this approach is that the
generation of genetically similar hosts (or genetically identical hosts, if immunosuppressants are used)
that display a range of disease outcomes provides a controlled system for comparing the effectiveness of
MCM candidates to protect against more severe disease. The use of genetic modification is largely limited
to the use of the mouse model system, for which there are a broad array of well-established tools.
However, the mouse model is less representative of human disease than other animal models, such as the
ferret. The use of immunosuppressants is a promising alternative. The key drawback of this approach is
that results gleaned from the use of immunocompromised hosts may not translate to disease in healthy
hosts.
The infection of wild type hosts with wild type viruses represents another alternative approach, which is
more relevant to human disease than other model systems. However, the utility of this approach for the
mouse model system is limited because mice are naturally resistant to infection with many wild type
influenza viruses. For the use of the ferret model system, wild type viruses may display a limited range of
virulence, which limits their utility for the development of MCMs that can protect against severe disease.
Moreover, the high genetic diversity among influenza viruses complicates the comparison of results from
the use of two genetically diverse wild type strains that exhibit varying levels of pathogenicity.
9.7.4.6 Benefits and Limitations of alt-GoF to Decision-making in Public Health Policy
Evaluation of the virulence of circulating animal influenza viruses detected through surveillance informs
assessment of their pandemic risk, which informs prioritization of investments in pre-pandemic
preparedness initiatives, such as pre-pandemic vaccine development. The contribution of alt-GoF
approaches to decision-making process in public health policy is discussed in detail in Sec. 9.6.4.3.
approaches
transmissibility relative to alt-GoF approaches are discussed, in each benefit category listed in the
NSABB Framework.
9.7.5.1 Scientific Knowledge Benefits GoF approaches relative to alt-GoF approaches
The underlying genetic and phenotypic features that result in infectivity, pathogenicity, and associated
morbidity and mortality during influenza virus infection are poorly understood, in part because of the
complex interplay between virus and host factors during pathogenesis. Because GoF and alt-GoF
approaches have distinct benefits and limitations for the study of viral factors versus host factors that
648 Pica N et al (2011) The DBA.2 mouse is susceptible to disease following infection with a broad, but limited, range of
influenza A and B viruses. Journal of virology 85: 12825-12829
649 Kim JI et al (2013) DBA/2 mouse as an animal model for anti-influenza drug efficacy evaluation. Journal of microbiology
(Seoul, Korea) 51: 866-871
650 van der Vries E et al (2013) Prolonged influenza virus shedding and emergence of antiviral resistance in
immunocompromised patients and ferrets. PLoS pathogens 9: e1003343
651 Belser JA et al (2011) The ferret as a model organism to study influenza A virus infection. Disease models & mechanisms 4:
575-579

contribute to pathogenicity, their relative value for identifying and characterizing virus factors versus host
factors is evaluated separately.
Identification and characterization of viral factors that contribute to pathogenicity
GoF approaches represent the most efficient and effective strategies for identifying novel viral genetic
traits that contribute to the pathogenicity of any virus strain. In addition, targeted genetic modification of
viruses to introduce traits associated with pathogenicity is uniquely capable of demonstrating that
particular viral genetic traits are necessary and sufficient to enhance virulence across multiple virus
contexts. However, results gleaned from cell culture and animal model studies may not translate to
humans. Notably, the use of attenuated strains for these studies is hindered by the fact attenuation may
alter disease pathogenesis, thus results may not be recapitulated in the genetic context of the wild type
virus. In addition, attenuated strains cannot be used when the mechanism of attenuation alters the viral
factor or underlying phenotype studied. However, the introduction of genetic traits associated with
virulence to lab-adapted strains provides a controlled system for the dissection of the functions of
individual genetic or phenotypic traits that contribute to virulence, and the fact that lab-adapted strains are
attenuated permits investigation of a large spectrum of virulence.
Although comparative sequence analysis of surveillance data has the potential to uncover viral genetic
traits that are associated with virulence in humans, the utility of this approach is significantly
compromised by shortcomings in the quality and availability of surveillance data. Additionally, this
approach is practically limited to the investigation of known viral genetic traits due to the high genetic
diversity among influenza viruses. For the same reason, characterization of wild type isolates is limited to
the study of previously known traits, unless genetically similar strains are available. In contrast,
comparative analysis of isolates within patients enables the identification of novel adaptive traits that are
associated with enhanced virulence over the course of infection. However this approach is often biased to
severe and late stage infection, which may not be representative. LoF approaches also have limited utility
for broad and unbiased identification of novel genetic and phenotypic traits due to their inefficiency,
including the fact that LoF approaches may uncover traits that indirectly contribute to pathogenicity.
Notably, targeted LoF enables the identification of genetic and phenotypic traits that are necessary for
enhanced virulence, which provides valuable information to complement and strengthen results gleaned
from targeted GoF studies.
While in vitro, virus free approaches and use of replication incompetent viruses enable the identification
of novel genetic and phenotypic traits that are necessary and sufficient to alter phenotypes underlying
pathogenicity, the importance of those genetic traits in the context of the complex host environment is
difficult to extrapolate. Moreover, the in vitro, virus free and cell culture methods do not provide any
information on mechanisms underlying the morbidity and mortality associated with influenza infection.
Finally, host-focused approaches provide indirect insight into the function of virus proteins and thus are
of limited utility for understanding how viral factors contribute to pathogenicity, relative to GoF
approaches.
Identification and characterization of host factors that contribute to pathogenicity
Both GoF and alt-GoF approaches can provide insight into host factors that enhance pathogenicity,
including deleterious immune responses that contribute to the morbidity and mortality caused by
influenza infection. GoF approaches can be used to identify host factors that are associated with enhanced
virulence and morbidity and mortality. In particular, targeted genetic modification to introduce traits that
are expected to enhance virulence provides a controlled system that can be used to tease apart the
interplay between virus and host factors contributing to pathogenesis, i.e. by demonstrating how changes

to a particular virus factor alter host immune responses and enhance infection-associated-pathology. The
utility of using risk-mediation reassortants in lieu of wild type viruses is significantly limited for the study
of host factors that contribute to pathogenicity due to differences in underlying pathogenesis mechanisms.
The main drawback of GoF approaches, with respect to the study of host factors that contribute to
pathogenicity, is that they cannot establish a causal link between a host factor and enhanced pathogenicity
and/or more severe disease pathology. Additionally, results from representative animal models may not
translate to humans.
The use of targeted knockout animals or pharmacological inhibition of the host factor during infection, an
alt-GoF approach, is uniquely capable of confirming that a host factor contributes to virulence and
pathogenicity. However, because the host response is dynamic and complex, inhibition of a host factor is
likely to have a multi-faceted effect on immune responses during infection, making the identification of
host traits that contribute to virulence difficult to resolve. Targeted genetic modification of viruses to
introduce traits expected to attenuate virulence (LoF) can also be used to identify host factors/responses
that are associated with enhanced pathogenicity. Like its GoF counterpart (i.e. targeted genetic
modification of viruses to introduce traits expected to enhance virulence), this approach provides a
controlled system for studying interplay between virus and host factors contributing to pathogenesis, and
the resulting information complements results from GoF studies. Immunological characterization of wild
type isolates exhibiting varied levels of virulence can demonstrate an association between a particular
host response and exacerbated disease pathology. However, this approach provides little mechanistic
insight into the role of particular virus-host interactions if viral isolates display high genetic diversity.
Several other alt-GoF approaches provide correlative data about the course of disease and the immune
responses that are associated with severe outcomes observed in humans, including comparative analysis
of genetic surveillance data, analysis of patient isolates, and analysis of autopsy data. This information is
highly valuable for connecting results observed in animal model systems to nature (e.g. whether
neurotropism observed during infections of ferrets with H5N1 viruses is representative of human
infections). However, these approaches provide limited mechanistic insight and are impaired by
limitations in the quality and availability of genetic surveillance data.
9.7.5.1.2 Scientific knowledge gap #2: Discover whether fitness defects associated with the acquisition of
GoF approaches are uniquely capable of proactively discovering compensatory mutations that rescue the
fitness of any antiviral-resistant strain with impaired growth, as well as establishing a causal link between
compensatory mutations and enhanced fitness. Computational modeling can be used to generate
hypotheses about mutations that may rescue growth, but all predictions must be experimentally confirmed
using GoF approaches. Comparative sequence analysis of antiviral-resistant strains with varied levels of
fitness has significant limitations relative to other approaches.
9.7.5.1.3 Scientific Knowledge benefit #3: Generation of animal models for the study of flu-associated
morbidity/mortality
observed during human infections are essential for the study of influenza-associated morbidity/mortality.
Although the ability to infect wild type hosts with wild type viruses would be ideal for translation of the
results of pathogenesis studies to human populations, mice are naturally resistant to infection with many
influenza viruses and/or wild type viruses may display a limited spectrum of disease in mice and ferrets.
In these cases, because pathogenicity and disease outcome is dependent on the interplay between virus
and host, both GoF and alt-GoF approaches enable the development of model systems that expand the
dynamic range of pathogenesis that is observed when using wild type viruses and wild type hosts. GoF
approaches achieve this goal by enhancing the virulence of the virus through serial passaging, while alt-

GoF approaches enhance host susceptibility to disease through targeted genetic modification or the use of
immunosuppressants. Both strategies generate animal models exhibiting a wider spectrum of disease than
observed in nature, which can be used to tease apart the relationship between host immune responses and
infection-associated morbidity and mortality. However, both GoF and alt-GoF approaches have
limitations. Serial passaging (GoF) may change the phenotypic properties of the virus in ways that alter
its biology, which could lead to misrepresentative findings. Modification of the host (alt-GoF) may alter
host immune responses that are involved in the response to infection, complicating translation of findings
to disease in healthy hosts. The genetic modification approach is limited to mice, although the use of
immunosuppressants represents a promising approach for ferrets, which are better representative of
human disease. Given these caveats, the use of model systems derived from GoF and alt-GoF approaches
strengthens the validity of any findings.
9.7.5.2 Surveillance Benefits of GoF approaches relative to alt-GoF approaches
The strategies for monitoring the virulence of circulating animal influenza viruses detected through
surveillance are similar to those for monitoring mammalian adaptation and transmissibility, and GoF and
alt-GoF approaches benefit surveillance through similar mechanisms. Thus, the relative benefits are
discussed collectively in Sec. 9.6.5.2.
9.7.5.3 Vaccine development Benefits of GoF relative to alt-GoF approaches
9.7.5.3.1 Vaccine development benefit #1: Development of new influenza vaccine candidates
A variety of vaccine platforms are being explored for the development of vaccines targeting avian
influenza viruses with pandemic potential. LAIVs have several characteristics that are desirable for
pandemic vaccines, but a major concern associated with their use is that the LAIV may recover virulence
upon growth in people. GoF approaches are uniquely capable of demonstrating whether LAIV strains
recover virulence upon growth in vivo, a critical aspect of vaccine safety testing prior to the conduct of
clinical trials. Other types of vaccines in development have strengths and weaknesses relative to LAIVs.
The type or types of vaccines that will ultimately prove to be most effective for avian influenza viruses is
not yet clear based on vaccinology research conducted to date. Given the need for effective pandemic
influenza vaccines, pursuing all promising strategies for vaccine development in tandem, including
LAIVs, will ensure that an effective vaccine is achieved in the shortest possible period of time.
9.7.5.3.2 Vaccine development benefit #2: targeted mutagenesis to remove virulence markers from
vaccine viruses
GoF approaches represent the most efficient and effective strategies for discovering novel genetic traits
that contribute to the virulence of influenza viruses. However, GoF approaches cannot be used to identify
or confirm genetic changes that are sufficient to attenuate the virulence of wild type strains, which is the
goal of vaccine virus development. LoF approaches, namely targeted mutagenesis, are uniquely capable
of identifying genetic changes (mutations or deletions) attenuate virulence across multiple virus strains.
Taken together, these approaches may enable the identification of novel virulence traits that can be
mutated to attenuate virulence, which can be applied to the production of AI vaccine viruses to further
improve the safety of the vaccine production process.
9.7.5.4 Therapeutic Development Benefits of GoF approaches relative to alt-GoF approaches
GoF approaches represent the most efficient and effective strategy for discovering novel viral virulence
factors that may be good therapeutic targets, but follow-up alt-GoF approaches are needed to confirm that
inhibiting the function of a particular viral factor is sufficient to attenuate or block viral replication and/or

reduce infection-associated pathology. Alt-GoF approaches are best-suited for discovering novel host
factors that contribute to virulence and immunopathology. However, GoF approaches can be used to gain
further mechanistic insight into the function of the host protein during infection, which strengthens the
evidence base for developing new therapeutics targeting that host factor. Two completely different
approaches for generating new therapeutic candidates are screening libraries of small molecule
compounds for their ability to inhibit viral replication in vitro and isolating monoclonal antibodies that
neutralize essential virus activities by directly binding to virus proteins, both of which have generated
promising therapeutic candidates that have entered clinical trials. Given that influenza viruses readily
acquire mutations that confer resistance to therapeutics and that different types of therapeutics may be
most effective against various influenza sub-types, a wide repertoire of therapeutics is needed to best
protect the public against the range of influenza threats that exist in nature. Pursuing all promising
pathways for therapeutic development in tandem, including GoF approaches, is the best strategy to
achieve this goal.
9.7.5.5 Benefits of GoF approaches relative to alt-GoF approaches for the development of both
vaccines and therapeutics
observed during human infections are essential for testing the safety and efficacy of new vaccines and
therapeutics. Although the ability to infect wild type hosts with wild type viruses would be ideal for
translation of the results of MCM development studies to human populations, mice are naturally resistant
to infection with many influenza viruses and/or wild type viruses may display a limited spectrum of
disease in mice and ferrets. In these cases, because pathogenicity and disease outcome is dependent on the
interplay between virus and host, both GoF and alt-GoF approaches enable the development of model
systems that expand the dynamic range of pathogenesis that is observed when using wild type viruses and
wild type hosts. GoF approaches achieve this goal by enhancing the virulence of the virus through serial
passaging, while alt-GoF approaches enhance host susceptibility to disease through targeted genetic
modification or the use of immunosuppressants. Both approaches provide a controlled system for
comparing the effectiveness of MCM candidates to protect against more severe disease, and both have
limitations. Serial passaging (GoF) may change the phenotypic properties of the virus in ways that alter
its susceptibility to the MCM in development, which would lead to misrepresentative findings.
Modification of the host (alt-GoF) may alter host immune responses that are involved in the mechanism
of action of the vaccine or therapeutic, complicating translation of findings to disease in healthy hosts.
The genetic modification approach is limited to mice, although the use of immunosuppressants represents
a promising approach for ferrets, which are better representative of human disease. Given these caveats,
the use of model systems derived from GoF and alt-GoF approaches strengthens the validity of any
findings.
9.7.5.6 Benefits to decision-making in public health policy
The relative contribution of GoF and alt-GoF approaches to benefit the decision-making process in public
health policy is discussed in detail in Sec. 9.6.5.3, as evaluation of the transmissibility of animal influenza
viruses similarly informs pandemic risk assessments and downstream decision-making.

9.8 Influenza viruses – Benefits of GoF research that leads to Evasion of Existing Natural
or Induced Adaptive Immunity
9.8.1 Summary
This section describes the benefits of GoF research that is reasonably anticipated to lead to evasion of
existing natural or induced adaptive immunity. Such GoF studies were found to generate scientific
knowledge, to inform surveillance of circulating seasonal influenza viruses, which has downstream
benefits to the production of seasonal influenza vaccines, and to benefit the development of new types of
influenza vaccines. Alt-GoF approaches that may generate similar benefits were also identified and
analyzed. At present, GoF studies resulting in evasion of existing natural or induced adaptive immunity
have unique benefits to scientific knowledge and surveillance, though full realization of GoF benefits to
surveillance requires scientific advancements and expansion of global public health surveillance
networks. Chapter 9.8 provides an overview of these benefits, including basic background and supporting
information; a fully referenced and more thorough discussion of these benefits can be found in Appendix
IV Sec. 15.5.
9.8.1.1 Benefits of GoF research to scientific knowledge
• GoF approaches:
o Are uniquely capable of providing in-depth information about the evolutionary mechanisms
driving antigenic drift as well as prospective information about currently circulating influenza
viruses. However, laboratory results may not translate to the evolution of flu viruses in
human populations.
o Are the most reliable and efficient method for discovering amino acid substitutions that
confer antigenic change to circulating viruses and are uniquely capable of demonstrating that
particular amino acid substitutions are necessary and sufficient to alter antigenicity. However,
these insights can be gleaned using attenuated 6:2 reassortant strains in lieu of wild type
viruses.
o Are the only method for mapping the antigenic sites of the HA protein in the context of the
full virus but are relatively low-throughput.
o Are uniquely capable of providing information about the antigenic evolution of influenza
viruses in nature but are constrained to studying the evolution of historic viruses in limited
depth.
o Allow for high-throughput mapping of antigenic sites using virus-free approaches, but results
may not be recapitulated in the context of the full virus.
9.8.1.2 Benefits of GoF research to surveillance
• GoF approaches:
o Are uniquely capable of strengthening the predictive value of molecular markers for antigenic
change, which can be used to infer phenotype from sequence. Use of molecular markers in
lieu of or to corroborate phenotypic testing results could improve the quality, timeliness, and
quantity of antigenic information about seasonal flu viruses detected through surveillance.

knowledge about the molecular basis of antigenic differences.
o Are critical for improving computational models for predicting antigenic phenotype based on
sequence. Use of computational models in lieu of or to corroborate phenotypic testing results
could improve the quality, timeliness, and quantity of antigenic information about seasonal
flu viruses detected through surveillance. However, the success of this approach is subject to
significant advancements in the accuracy of existing models.
o Full realization of these GoF benefits will require expansion of sequencing capabilities
diagnostic labs involved in global influenza surveillance.
o Have significant limitations for strengthening the predictive value of molecular markers for
antigenic changes.
o Are also critical for improving computational models for predicting antigenic phenotype
based on sequence, but through the generation of different types of data that complement data
generated through GoF approaches.
o Phenotypic assays for antigenic characterization are uniquely capable of providing direct
information about antigenicity, but results may be delayed relative to the publication of viral
sequences.
9.8.1.3 Benefits of GoF research to vaccine development
• GoF approaches:
o GoF approaches that improve sequence-based predictions of antigenicity have potential to
increase the robustness, quantity, and timeliness of antigenic characterization data upon
which strain selection decisions are based. However, full realization of this benefit depends
on the expansion of sequencing capabilities at National Influenza Centres.
o GoF approaches have potential to improve the ability to predict antigenic drift, through
experimental and/or computational methods, which would allow the production of
“antigenically advanced” vaccines that match circulating strains at their time of deployment.
knowledge about the evolutionary mechanisms driving antigenic drift.
o Are uniquely capable of defining the antigenic landscape of the HA protein (the spectrum of
antigenic configurations that HA can assume and which regions of HA can mutate while
preserving virus viability). These data may inform the development of broad-spectrum or
universal flu vaccines.
o Efforts to improve antigenic characterization assays, in order to improve the quality of
antigenic characterization data upon which strain selection decisions are based, are ongoing
but have had limited success to date.
o Strengthening global influenza surveillance networks will improve the quantity, timeliness,
and representativeness of data upon which strain selection decisions are based, but these
efforts face considerable funding and political barriers.

o Alternative strategies for the development of broad-spectrum or universal flu vaccines are
being pursued and have also shown promise.
9.8.2 Overview of GoF research landscape: evasion of existing natural or induced adaptive
immunity
9.8.2.1 Serial passaging of viruses in the presence of cognate antibodies
Serial passaging of viruses in the presence of cognate antibodies may lead to the acquisition of mutations
that allow the virus to escape neutralization by the antibody. This experiment can be performed in cell
culture using monoclonal antibodies, convalescent sera from infected individuals, post-infection ferret
sera, or in animals that have been vaccinated or previously exposed to influenza viruses. Sequencing of
emergent antibody escape viruses identifies amino acid substitutions that are sufficient to confer antigenic
change, which provides a foundation for follow-up studies investigating the molecular basis of antigenic
differences between strains. Additionally, sequencing viral isolates at multiple stages of the selection
process and determining the effect of amino acid substitutions on viral fitness and other virus phenotypes
provides insight into the evolutionary mechanisms driving antigenic drift. Finally, when performed in
vitro using monoclonal antibodies, the location of escape mutations reveals potential antibody epitope
sites.
9.8.2.2 Forward genetic screen to identify mutations that alter antigenicity
Forward genetic screens involve random mutagenesis of the HA protein followed by characterization of
the antigenicity of mutants using the hemagglutination inhibition (HAI) assay or other assays, in order to
identify amino acid substitutions that do and do not lead to antigenic change. Follow-up studies may
determine the consequences of antigenicity-altering mutations on other virus phenotypes, such as viral
fitness and pathogenicity. As for serial passaging experiments, the identification of amino acid
substitutions that confer antigenic change provides a foundation for studies investigating the molecular
basis of antigenic differences. In addition, comprehensive forward genetic screens can be used to define
the ‘antigenic landscape’ of the HA protein – that is, which substitutions the HA protein will tolerate and
which of those substitutions cause antigenic drift.
9.8.2.3 Targeted modification of viruses to introduce mutations that are expected to alter antigenicity
A final GoF approach that may lead to viruses that evade existing adaptive immunity involves targeted
genetic modification to introduce mutations that are expected to alter antigenicity, followed by antigenic
characterization of the mutant virus using the HAI assay or other assays. Of note, mutations may be
identified through GoF approaches, such as serial passaging of viruses in the presence of cognate
antibodies, or alt-GoF approaches, such as comparative analysis of historical sequences. This approach
demonstrates that a particular mutation or set of mutations is necessary and sufficient to alter antigenicity,
which provides a foundation for follow-up studies investigating the molecular basis of antigenic
differences between strains.
When any of these three approaches are performed using recently or currently circulating seasonal
influenza viruses or using seasonal viruses that have recently served as the basis for vaccine strains, the
end result is the generation of a mutant strain that cannot be neutralized by existing natural or induced
immunity, respectively. Because human populations do not have widespread immunity to the 1918 H1N1

pandemic virus or to animal influenza viruses (i.e. avian viruses652 and swine viruses653), no approaches
involving these viruses meet this phenotypic criterion. Therefore, this section does not include studies that
investigate the mechanisms underlying antigenic drift of avian strains in response to selection pressure
from vaccination or the chicken immune system, nor any other studies focused on animal influenza
strains. Note that because these studies may lead to the acquisition of mutations in the influenza HA
protein, which is a critical determinant of mammalian adaptation, transmissibility, and virulence, these
studies may result in the generation of viruses with altered virulence, infectivity, and transmissibility from
a “human” perspective. However, whether and what phenotypic changes are likely to arise cannot be
anticipated with certainty.
Finally, GoF approaches may also lead to the generation of influenza viruses that are capable of evading
recognition by the host innate immune system. Because virus interactions with innate immune factors are
critical determinants of virulence, these approaches are evaluated in the “enhanced morbidity and
mortality in appropriate animal models” section.
In this section, the ability of GoF methods to address three unanswered questions in this field are
evaluated:
• How do influenza viruses evolve antigenically in response to immune pressure? That is, what are
the evolutionary mechanisms driving antigenic drift, including the role of different selection
pressures (e.g. vaccination) and the interplay between antigenic escape and other virus
phenotypes, such as fitness?
• What is the molecular basis of antigenic drift? That is, what amino acid substitutions in the HA
protein lead to antigenic change, and what is the biophysical basis of that effect?
• What are the antigenic sites on the HA protein that are targeted by human monoclonal antibodies?
Influenza viruses circulating in nature acquire mutations in response to immune pressure from human
populations that allow the viruses to escape recognition by the adaptive immune system, a process termed
“antigenic drift”.654 As a result, the strain composition of the seasonal influenza vaccine must be updated
annually to ensure that the vaccine strains antigenically “match” circulating strains. Research in this area
is focused on the influenza HA protein, which is the immunodominant influenza protein and represents
the primary component of current influenza vaccines. The mechanisms underlying antigenic drift of the
HA protein and the relationship between genotype and antigenic phenotype are not well understood. One
of the knowledge gaps that contributes to this uncertainty is an incomplete understanding of the antigenic
sites on the HA protein that are targeted by human monoclonal antibodies, as these sites are presumably
hotspots for antigenic evolution.655
652 Jernigan DB, Cox NJ (2015) H7N9: Preparing for the Unexpected in Influenza. Annual Review of Medicine 66: 361-371
653 Skowronski DM et al (2012) Cross-reactive and vaccine-induced antibody to an emerging swine-origin variant of influenza
A virus subtype H3N2 (H3N2v). J Infect Dis 206: 1852-1861
654 Webster RG et al (1982) Molecular mechanisms of variation in influenza viruses. Nature 296: 115-121
655 O'Donnell CD et al (2012) Antibody pressure by a human monoclonal antibody targeting the 2009 pandemic H1N1 virus
hemagglutinin drives the emergence of a virus with increased virulence in mice. MBio 3

9.8.3.1.1 Scientific Knowledge Gap #1 – How do influenza viruses evolve antigenically in response to
immune pressure?
GoF approaches that involve serial passaging of viruses in the presence of cognate antibodies provide
insight into the evolutionary mechanisms driving antigenic drift in response to immune pressure. Both in
vivo and in vitro approaches have unique strengths. Namely, subjecting viruses to selection from the full
complement of the animal immune system better mimics the selective pressure viruses experience in
humans, while in vitro approaches can be conducted using convalescent sera (or isolated antibodies) from
people, which may be more relevant to humans than selective pressures in animals. In addition, the in vivo
approach represents a controlled system for studying the role of selective pressures from prior exposure to
influenza viruses through natural infection and/or vaccination in shaping antigenic evolution. In both
cases, results from laboratory studies may not translate to the evolution of viruses in human populations
in nature and may not be conserved in other virus contexts. Importantly, follow-up studies can determine
the effect of antigenic drift on other virus phenotypes, such as fitness, which provides insight into how
likely mutations are to persist in a host or in a population once they have arisen.
9.8.3.1.2 Scientific Knowledge Gap #2 – What is the molecular basis of antigenic drift?
Several GoF approaches can be used to discover mutations that lead to antigenic drift, which provides a
foundation for follow-up studies investigating the biophysical basis of antigenic change. First, serial
passaging of viruses in cells in the presence of cognate sera or monoclonal antibodies, or in animals that
have been vaccinated or previously exposed to influenza viruses, leads to the emergence of antigenic
escape mutants. Sequencing the HA gene of emergent escape viruses reveals mutations that are sufficient
to alter virus antigenicity. This approach is highly efficient and can be applied to any virus, including
currently circulating strains. Notably, in vitro and in vivo selection approaches equally enable the
identification of mutations associated with antigenic drift, though the in vitro approach is faster and
cheaper. Importantly, as multiple mutations may arise during passaging, follow-up studies may be needed
to determine which mutation(s) are responsible for the antigenic escape phenotype.
Forward genetic screens, which involve mutagenesis of the HA protein and subsequent characterization of
the antigenicity of mutant viruses, represent another GoF approach for identifying mutations that confer
antigenic change. Though screening for escape mutants is more labor-intensive than selection methods
based on serial passaging, the screening approach is uniquely capable of identifying mutations that do not
lead to antigenic change, which critically informs efforts to develop models for the sequence-based
prediction of antigenicity. Importantly, because of the influence of genetic context on antigenicity,
antigenic escape mutations identified through either serial passaging or forward genetic screens may not
generalize to other virus strains within the same or different HA subtype.
Finally, targeted genetic modification of viruses to introduce mutations associated with antigenic change,
followed by antigenic characterization of mutant viruses, is used to demonstrate that mutations are
necessary and sufficient to alter antigenicity. Subsequently, to determine whether the phenotypic
consequences of mutations are functionally generalizable across multiple virus strains, targeted
mutagenesis can be used to introduce mutations into new virus strains, followed by antigenic
characterization. Together, these results provide a strong foundation for follow-up structural studies to
determine the biophysical basis of antigenic differences and critically inform the development of models
for the prediction of antigenic phenotype from genotype.

9.8.3.1.3 Scientific Knowledge Gap #3 – What are the antigenic sites on the HA protein that are targeted
by human monoclonal antibodies?
Serial passaging of viruses in cells in the presence of monoclonal antibodies to select for antibody escape
mutants is a classic method for identifying putative antibody binding sites. Specifically, the amino acid
positions where mutations arise represent potential antigenic sites, although interpretation of this data is
complicated by the fact that mutations outside antibody binding sites can impact HA-antibody
interactions through long-range effects. In the event that multiple mutations arise within the HA protein,
targeted mutagenesis to introduce individual mutations into the parental strain may be used to confirm
which mutations are necessary and sufficient to confer escape. This approach is simple, rapid, and allows
for precise mapping of antigenic sites. However, each passaging experiment focuses on the identification
of a single antigenic site, such that multiple rounds of passaging with distinct antibodies are required to
map multiple antigenic regions.
9.8.3.2 Benefits and limitations of GoF approaches to surveillance
GoF approaches that lead to the identification of mutations that alter antigenicity have potential to aid
antigenic surveillance of human seasonal influenza viruses by facilitating prediction of antigenic
phenotype from genotype, in lieu of isolating and experimentally evaluating the antigenicity of viruses.
Specifically, GoF data can strengthen the predictive value of molecular markers for antigenic change and
can improve models for predicting antigenic phenotype from genotype. Either application has the
potential to aid the bi-annual selection of strains for the seasonal influenza vaccine, as described in the
“informing policy decisions” section below.
9.8.3.2.1 Introduction to influenza virus surveillance: current practices and limitations
The WHO Global Influenza Surveillance and Response System (GISRS) conducts surveillance of
seasonal influenza viruses year-round. The major goal of seasonal flu surveillance is to monitor the
antigenic evolution of viruses – that is, to detect when new antigenic variants emerge in human
populations and to determine their prevalence and geographic distribution.656,657 A global network of
National Influenza Centres (NICs) collect clinical specimens in their countries and ship viral isolates to
one of six WHO Collaborating Centres (WHOCCs) for detailed antigenic characterization.658,659 These
data critically inform WHO-coordinated decisions about which strains to recommend including in the
seasonal flu vaccine, which are developed during bi-annual Vaccine Composition Meetings (VCMs).660,661
If surveillance data indicate that a new antigenic variant has emerged and spread geographically, the
WHO strain selection committee will recommend updating that component of the vaccine.
656 Ampofo WK et al (2013a) Strengthening the influenza vaccine virus selection and development process: outcome of the 2nd
657 Ampofo WK et al (2015) Strengthening the influenza vaccine virus selection and development process: Report of the 3rd
WHO Informal Consultation for Improving Influenza Vaccine Virus Selection held at WHO headquarters, Geneva,
Switzerland, 1-3 April 2014. Ibid. 33: 4368-4382

Antigenic characterization primarily relies on the hemagglutination inhibition (HAI) assay developed in
the 1940s.662 Though simple and inexpensive, HAI assays have several significant drawbacks that
compromise their utility and reliability for antigenic characterization.663,664 GoF approaches have potential
to address this shortcoming by improving two methods for predicting antigenic phenotype based on
sequence, thereby improving antigenic surveillance of seasonal influenza viruses. First, HA sequences
can be inspected for the presence or absence of molecular markers for antigenic drift that were identified
through GoF approaches. Second, that same GoF-derived data can be used to improve existing models for
predicting antigenicity based on genotype. In either case, that information could supplement phenotypic
characterization data, to strengthen the certainty of conclusions about antigenic relationships between
strains, or could be used in lieu of phenotypic characterization data.
9.8.3.2.2 Analysis of GoF approaches that support the use of molecular markers to evaluate the
antigenicity of seasonal influenza viruses
During the current strain selection process, HA sequences are inspected for the presence of amino acid
substitutions that are known to be associated with altered antigenicity. This information can be used to
corroborate antigenic characterization data from the HAI assay or can help to resolve antigenicity
questions when HAI assay results are difficult to interpret. While this information informs the decision-
making process, the utility of these markers is limited by significant uncertainties in the state of this
science. First, the ability to reliably predict whether a particular amino acid substitution will confer
antigenic change in a new genetic context is poor. Second, because other, as-yet-undiscovered amino acid
changes may alter antigenicity, the absence of known markers is not yet meaningful (i.e. does not indicate
that the antigenicity of the strain is unchanged).
GoF approaches are critical for addressing both aspects of scientific uncertainty described above to
strengthen the utility of molecular marker data for antigenic change. To strengthen the predictive value of
molecular markers for antigenic change, several types of experiments are needed:
• Targeted mutagenesis to introduce known genetic markers for altered antigenicity into new
genetic contexts (i.e. validate the antigenic consequences of the marker in a variety of strain
contexts), which represents a GoF approach,
• Targeted mutagenesis to determine which amino acid substitutions at a particular site previously
associated with antigenic change are sufficient to alter antigenicity, which represents a GoF
approach, and
• Experiments that explore the antigenic plasticity of the HA protein, to discover new substitutions
that confer antigenic change as well as substitutions that do not alter antigenicity.
To address the third experimental goal, two GoF approaches (serial passaging and forward genetic
screens) are capable of uncovering novel mutations that confer antigenic change, and targeted
mutagenesis can be used to confirm their causality (also GoF). Although these data will undoubtedly
strengthen the predictive value of molecular markers for antigenic change, given the importance of
662 Hirst GK (1942) THE QUANTITATIVE DETERMINATION OF INFLUENZA VIRUS AND ANTIBODIES BY MEANS
OF RED CELL AGGLUTINATION. J Exp Med 75: 49-64

genetic context on influenza biology, significant challenges face any effort to improve the predictive
value of such markers to a level that is meaningful. Whether this goal is achievable will depend on
whether the number of amino acid substitutions that HA can accept is limited or very large, which is as-
yet-unknown. In addition, the fact that negative results are generally not published in the scientific
literature also hinders advancements in this area, as knowing when markers are not conserved critically
informs their utility.
9.8.3.2.3 Analysis of GoF approaches that improve predictive models
GoF data can also be used to improve the quality of computational models for predicting antigenic
phenotype from genotype, which represents a different sequence-based approach for predicting
antigenicity. Current models cannot accurately predict antigenic phenotype from genotype.665
GoF approaches have potential to improve these models in two ways: (1) by generating experimental data
about novel antigenic changes that are necessary and sufficient to alter antigenicity, which can be
incorporated into datasets used to train the models, and (2) by testing predictions of novel mutations that
would affect antigenicity that these models make, the results from which will feed back to improve model
accuracy. As existing models are primarily trained using historical data (i.e. the sequences and antigenic
characterization data from historical isolates), the ability of GoF approaches to explore new antigenic
space will complement existing data sources to enhance the predictive capability of these models for
currently circulating isolates that are evolving antigenically in new ways. As above, the feasibility of
developing models that can accurately predict antigenic phenotype from genotype will depend on the
antigenic plasticity of the HA protein, which is currently unknown.
If the landscape of amino acid substitutions that can give rise to antigenic change is large, then molecular
markers and computational models may never be robust enough to replace antigenic characterization data
generated through laboratory assays. Nonetheless, given the shortcomings of phenotypic assays for
characterizing antigenicity, the ability to corroborate laboratory results using sequence-based predictions
can significantly strengthen the quality of antigenic characterization data, particularly if clinical
specimens are directly sequenced.
9.8.3.3 Benefits and Limitations of GoF approaches to vaccine development
9.8.3.3.1 Vaccine development benefit 1: Improve strain selection capabilities for seasonal influenza
vaccines
GoF approaches have potential to improve the strain selection process for seasonal influenza vaccines in
several ways. First, a critical factor in strain selection is analysis of the antigenic characteristics of
circulating influenza viruses, to determine whether new antigenic variants have emerged. As described in
Sec. 9.8.3.2, GoF data can improve methods for predicting antigenic phenotype from genotype, which
may provide several advantages over the use of traditional, laboratory-based antigenic characterization
methods. In addition, GoF approaches have the potential to aid efforts to predict antigenic drift, either
directly through the selection and analysis of drifted strains or by informing the development of models
for predicting drift. As selected strains sometimes drift during the course of vaccine development, which
leads to poor vaccine match, these efforts could improve the efficacy of vaccines by enabling deliberate
production of “drifted” strains that match circulating strains at the time of vaccine deployment.

Introduction to strain selection for seasonal flu vaccines: current practice and limitations
Since the early 1970s, the WHO has provided formal recommendations for the strain composition of
seasonal influenza vaccines based on year-round influenza surveillance conducted through the GISRS
(described above).666,667 Experts must predict which strains are likely to be dominant six to eight months
in advance of the start of the target flu season, to provide sufficient time for manufacturing the
vaccine.668,669 Despite the complexity of the data considered and the challenge of predicting dominant
strains many months in advance, this process generally works well - most years, the vaccine is well-
matched to circulating strains.670 However, occasionally a rare antigenic variant rises to prominence
during the course of vaccine production, as happened during the recent 2014 – 2015 flu season for the
H3N2 strain, which results in poor vaccine match and reduced vaccine efficacy.671,672
Several shortcomings compromise the efficacy of the current strain selection process. First, the timeliness
and representativeness of isolates forwarded to WHOCCs by NICs, which form the basis of strain
selection recommendations, could be improved. In particular, due to significant lag times between sample
collection and shipment (e.g. two to three months between 2010 and 2012 in the WHOCC London
region), many isolates cannot be analyzed in time for consideration during VCM meetings. These
shortcomings in existing surveillance networks reduce the quality and quantity of input data for strain
selection decisions, which compromises the accuracy of the process. A second shortcoming of the current
strain selection process is its heavy reliance on the HAI assay for antigenic characterization of
surveillance isolates, which suffers several significant drawbacks. A final shortcoming is the inability to
reliably predict whether rare antigenic variants will rise to prominence in nature during the vaccine
production process, which results in poor vaccine match.
GoF approaches that lead to evasion of existing natural or induced immunity have potential to address all
three shortcomings in the current strain selection process, through several different mechanisms.
Analysis of GoF approaches that improve strain selection capabilities by improving antigenic surveillance
As discussed above, GoF approaches have potential to strengthen the predictive value of molecular
markers for antigenic drift and to improve the accuracy of existing models for predicting antigenic
phenotype from genotype. Either strategy for sequence-based prediction of antigenic phenotype could be
used to corroborate lab-generated HAI data in cases where results are difficult to interpret, thereby
improving the quality of input data for the strain selection decision. Alternatively, sequence-based
prediction methods could replace laboratory methods for antigenic characterization. Given that sequence
data can be collected rapidly and economically and is increasingly being generated at NIC labs, reliance
666 Oshitani H (2010) Influenza surveillance and control in the Western Pacific Region. Western Pacific surveillance and
response journal : WPSAR 1: 3-4
667 WHO. Process of Influenza Vaccine Virus Selection and Development
http://apps.who.int/gb/pip/pdf_files/Fluvaccvirusselection.pdf. Last Update November 19, 2007. Accessed November 22,
2015.
670 Legrand J et al (2006) Real-time monitoring of the influenza vaccine field effectiveness. Ibid. 24: 6605-6611
Conferences (CICG) Geneva, Switzerland, 7 to 9 December 2011. Ibid. 31: 3209-3221
672 Xie H et al (2015) H3N2 Mismatch of 2014-15 Northern Hemisphere Influenza Vaccines and Head-to-head Comparison
between Human and Ferret Antisera derived Antigenic Maps. Sci Rep 5: 15279

on sequence data may allow for consideration of a greater number of isolates, including isolates collected
close to the VCM meeting dates. The result, an increase in the quantity of input data for the strain
selection decision, would improve the process through a different mechanism. Critically, although
molecular marker data informs strain selection decisions, neither molecular marker data nor predictive
models are currently robust enough to replace phenotypic data (and may never be). Notably, GoF
approaches are uniquely critical for advancing the state of the science for both approaches. Finally, full
realization of these benefits requires continued expansion of sequencing capabilities at NICs, as only
about one-quarter to one-half of HA sequences for seasonal flu strains are currently generated at NICs
(depending on the influenza sub-type).673
Analysis of GoF approaches that improve strain selection capabilities through prediction of antigenic drift
GoF approaches to experimentally induce drift can be used to predict how circulating viruses may drift in
nature, enabling production of vaccines against future, “drifted” strains that will antigenically match
circulating viruses at their time of deployment. Specifically, the selection of antibody escape mutants of
currently circulating viruses, through serial passaging or forward genetic screens conducted in vitro and in
vivo, enables the identification of HA substitutions that confer escape. Coupled with genetic surveillance
data, this information can be used to forecast the antigenicity of the next dominant strain to arise in
nature.674,675 However, whether and when such variants will emerge is uncertain, in part because
stochastic events in natural evolution may result in the appearance of an unusual mutant that was not
selected in the experimental studies. For that reason, this data is not currently incorporated into the strain
selection process, and additional research is needed to determine whether it will be useful for predicting
the course and timing of antigenic evolution in the future.676
A different approach for predicting antigenic drift involves the use of computational models for antigenic
evolution (though computational models could be used in conjunction with experimental data). Existing
models cannot reliably predict antigenic drift, and two types of GoF studies are needed to improve the
quality of existing models. First, a better understanding of the process of antigenic evolution will provide
a foundation for the design of better models. As described above (Sec. 9.8.3.1.1), GoF approaches are
uniquely capable of providing in-depth information about the evolutionary mechanisms driving antigenic
drift as well as prospective information about the evolution of currently circulating viruses. Second,
influenza modeling experts have stated that developing the ability to predict whether particular amino
acid substitutions alter antigenicity in a given genetic context is critical for advancing the quality of these
models.677,678 As described in the preceding section, GoF approaches are essential for improving the
accuracy of models for prediction of antigenic phenotype from genotype, although other types of data are
also needed.
Taken together, utilizing experimental and/or in silico approaches to predict whether new antigenic
variants are likely to emerge during the course of vaccine production would enable the production of
vaccines based on those predicted future strains. This strategy would increase the likelihood that vaccines
673 (2015w) Personal communication from WHOCC representative.


match the strains that are circulating during their target flu season, which will lead to an overall
improvement in vaccine efficacy. One key concern associated with this strategy is that evolutionary
predictions are difficult and are unlikely to be correct one hundred percent of the time, even as the science
of prediction advances. Importantly, the exact amino acid sequence of the next dominant strain does not
need to be predicted, but rather its antigenicity (as multiple sequences can fall into the same antigenic
“cluster”). In addition, studies have shown that immunization with “antigenically advanced” vaccines, i.e.
those that are based on predicted future strains, can provide some degree of protection against currently
circulating strains.679 Thus, even if the prediction is incorrect (i.e. the strain does not drift in nature), pre-
emptive vaccination strategies are likely afford some degree of protection.
9.8.3.3.2 Vaccine development benefit #2: Development of broad-spectrum or universal flu vaccines
Researchers are actively pursuing the development of broad-spectrum flu vaccines, which could protect
against multiple strains (a subset of related strains within a subtype, an entire subtype, or multiple
subtypes), and “universal” flu vaccines, which could protect against all strains. Either type of vaccine
would eliminate the need for an exact match between vaccine strains and circulating seasonal viruses,
thus improving the efficacy of seasonal flu vaccines. In addition, universal or broad-spectrum vaccines
could be available rapidly during a pandemic or could be used to pre-vaccinate the population against
emerging influenza strains, thereby increasing vaccine coverage during a pandemic. Scientists are
exploring multiple strategies for development of such next-generation influenza vaccines, and both GoF
and alt-GoF approaches have potential to inform this process.
Analysis of GoF approaches that inform the development of broad-spectrum or universal flu vaccines
GoF approaches that aim to map the antigenic landscape of the HA protein have potential to inform the
development of broad-spectrum and universal influenza vaccines. Specifically, comprehensive forward
genetic screens to identify which substitutions the HA protein can tolerate and which of those
substitutions alter antigenicity will define the regions of the HA protein could drift (i.e. without
significantly compromising the stability of HA and the viability of the virus) as well as how those regions
can change antigenically. Defining all possible antigenic configurations of the HA protein provides a
foundation for developing a broad-spectrum vaccine (or vaccine cocktail) that protects against a large
fraction of the possible antigenic space, thus pre-empting antigenic drift in nature and eliminating the
need for annual production of seasonal flu vaccines.680 Alternatively, defining those regions of the HA
protein that do not mutate may provide a foundation for the development of a “drift-resistant” universal
vaccine that targets those regions. Currently, whether either strategy will lead to the development of an
effective influenza vaccine is unknown. In addition, comprehensive mapping of the antigenic landscape
represents a labor-intensive, long-term project, and whether findings will be specific to an influenza strain
or sub-type or will translate to other virus strains is unknown.
9.8.3.4 Benefits and limitations of GoF approaches to the development of therapeutics and diagnostics
GoF approaches in this phenotypic category are focused on elucidating mechanisms of antigenic drift in
response to immune pressure, which is not relevant for the development of therapeutics or diagnostics.
(We note that studies that generate escape mutants from candidate monoclonal antibody therapeutics,
which are experimentally similar to approaches described above, are discussed in the “evasion of
therapeutics” section.)
679 Fonville JM et al (2014) Antibody landscapes after influenza virus infection or vaccination. Science 346: 996-1000
680 (2015u) Interview with Biomedical Advanced Research and Development Authority representative.

9.8.3.5 Benefits and limitations of GoF approaches to policy decisions
GoF approaches have potential to inform the selection of strains for the seasonal influenza vaccine in
several ways, as described above.
GoF approaches that inform strain selection for seasonal influenza vaccines may improve the efficacy of
seasonal flu vaccines by increasing the likelihood that the vaccine strains will match the strains that are
circulating during the target influenza season. Ultimately, this benefit may increase vaccine uptake but
otherwise is unlikely to yield economic benefits.
9.8.4.1 Benefits and limitations of alt-GoF approaches to Scientific Knowledge
immune pressure?
The use of attenuated strains for serial passaging studies, in lieu of wild type strains, represents one type
of alt-GoF approach for the study of antigenic evolution. Two types of attenuated strains are used for
serial passaging studies to investigate antigenic evolution mechanisms: the mouse-adapted strain PR8,
which is avirulent in people,681 and 6:2R strains that contain the HA and NA gene segments from a
seasonal strain of interest and the remaining six gene segments from PR8. While use of either type of
attenuated strain can provide insight into the basic mechanisms of antigenic evolution, results may not
translate to wild type strains due to differences in disease pathogenesis caused by wildtype versus
attenuated strains as well as other factors. Moreover, 6:2R strains cannot be used to predict the effect of
antigenic escape mutations on the fitness of wildtype strains because in vivo fitness is a complex, multi-
genic trait that is highly dependent on genetic context. Finally, as the PR8 strain and 6:2R strains do not
efficiently infect ferrets,682 these studies are limited to the use of mouse model systems, which is less
representative of human disease than the ferret model system.
Comparative analysis of historical virus sequences that have drifted antigenically over time represents
another alt-GoF approach for studying antigenic evolution. Relative to GoF approaches, the strength of
the comparative sequence analysis approach is that it provides insight into the antigenic evolution of a
wide breadth of influenza viruses in human populations. However, the success of this approach depends
on the quality of available surveillance data; some strains have limited numbers of sequences available,
and biases in the way that some surveillance data are collected render the data unsuitable or difficult to
use. An additional limitation is that the historical record is static – that is, it cannot provide insight into
mutations that were selected against, which is important knowledge for understanding the pressures and
constraints that guide antigenic evolution. Finally, this approach cannot be used to proactively study the
antigenic evolution of currently circulating viruses.
In silico approaches can be also used to investigate mechanisms underlying antigenic drift of influenza
viruses. Existing models are largely based on and have been validated using historical data. As a result,
the quality of these models is constrained by the set of limitations described above for the comparative
681 Beare AS et al (1975) Trials in man with live recombinants made from A/PR/8/34 (H0 N1) and wild H3 N2 influenza
viruses. Lancet 2: 729-732
682 Jin H et al (2004) Imparting Temperature Sensitivity and Attenuation in Ferrets to A/Puerto Rico/8/34 Influenza Virus by
Transferring the Genetic Signature for Temperature Sensitivity from Cold-Adapted A/Ann Arbor/6/60. J Virol 78: 995-998

sequence analysis approach. Although models can provide insight into the relationships between genetic
and antigenic evolution, their accuracy in predicting future antigenic drift is unknown, thus any
predictions must be experimentally validated.
The use of attenuated reassortant strains containing the HA and NA genes from a seasonal strain of
interest and the remaining six “internal” genes from the lab-adapted, attenuated strain PR8 (6:2R strains)
in lieu of wild type strains represents one type of alternative approach for the study of the molecular basis
of antigenic drift. Because the antigenicity of the HA protein is preserved in the context of a 6:2R
strain,683 6:2R strains are as suitable as wild type strains for the discovery and confirmation of amino acid
substitutions that lead to antigenic drift using in vitro or mouse model systems.
Several alternative experimental approaches can also be used to identify mutations associated with
antigenic change. Comparatively analyzing the sequences of natural isolates that have drifted
antigenically over time can lead to the identification of mutations that are associated with antigenic
change. However, follow-up GoF experiments are needed to establish a causative link between particular
mutations and antigenic change. Another drawback of this approach is that it is limited to the
identification of amino acid substitutions that have arisen in nature, which represents a fraction of the
possible antigenic space.
In silico approaches represent another alt-GoF approach for the identification of mutations associated
with antigenic drift. Specifically, computational models based on antigenic, sequence, and HA structural
data can be used to predict amino acid substitutions that will alter antigenicity. Although computational
approaches can fully explore all possible antigenic configurations, existing models cannot predict
mutations that will lead to antigenic change with certainty, thus the phenotypic consequences of any
predicted mutation must be confirmed experimentally.684
by human monoclonal antibodies?
Several alt-GoF approaches can also be used to map the antigenic epitopes of the influenza HA protein.
One approach involves the use of cell surface display systems in yeast, bacteria, or bacteriophages. These
systems exploit the ability of these organisms to express random peptides or protein fragments from the
HA protein on their cell surface. Libraries of mutant bacteria/phages/yeast can then be screened for
binding to a monoclonal antibody or post-infection sera, for mapping of the antigenic epitope of a
particular antibody or comprehensive mapping of antigenic sites, respectively. The main strength of this
approach is that it is high-throughput, allowing for mapping of multiple antigenic sites at once through the
use of complex sera or multiple mAbs. However, as the presentation of mapped epitopes may be different
in the context of the full virus, GoF experiments with full virus should be performed to validate any
findings.
Another alternative approach involves analysis of crystal structures of a viral protein (or protein fragment)
complexed with a particular mAb. The crystal structure demonstrates precisely where an antibody binds
to the HA protein, which can be compared to previous studies to determine whether the epitope is
previously known or novel. The main drawback of this approach is that it is labor- and time-intensive and
therefore has limited throughput. Additionally, researchers have faced technical limitations, such as

684 (2015x) Influenza strain selection. Interview with industry personnel involved in vaccine production. .

difficulty crystallizing full-length HA proteins and radiation damage during the data collection process,
which may compromise the quality of the data.685
9.8.4.2 Benefits and limitations of alt-GoF approaches to surveillance
As described above, GoF approaches have the potential to benefit antigenic surveillance for human
seasonal influenza viruses in two ways: (1) by improving the predictive value of molecular markers for
antigenic drift, and (2) by improving the accuracy of models for predicting antigenic phenotype from
genotype. This section evaluates the ability of alternative experimental approaches to similarly strengthen
the utility of molecular marker data and predictive models to understand whether alt-GoF approaches
have the potential to benefit surveillance through either mechanism.
9.8.4.2.1 Analysis of alt-GoF approaches that support the use of molecular markers to evaluate the
antigenicity of seasonal influenza viruses
Currently, the predictive value of molecular markers for antigenic drift is limited by three sources of
scientific uncertainty: (1) whether markers alter antigenicity in different genetic contexts, (2) whether
novel amino acid substitutions at particular sites that are known to be associated with antigenic drift will
alter antigenicity, and (3) what other amino acid substitutions confer antigenic change. Characterizing the
antigenicity of wild type viruses that contain known molecular markers can demonstrate whether a known
marker is associated with altered antigenicity in a new genetic context, but no alt-GoF approaches are
capable of validating that the marker is necessary and sufficient to confer antigenic change in a new
strain, which is essential for application of that knowledge to surveillance.686 Similarly, characterization
of wild type viruses is limited to determining whether different mutations at known sites or novel
mutations are associated with antigenic change. Given the limited accuracy of existing models,
predictions of any type must be experimentally confirmed using GoF approaches. However, in all cases,
attenuated reassortant strains can be used in lieu of wild type strains because the antigenicity of the 6:2R
strain is similar to that of the parental wild type strain.
9.8.4.2.2 Analysis of alt-GoF approaches that can improve predictive models
Existing models for prediction of antigenic phenotype from genotype are largely built and validated using
historical data. Though comparative analysis of additional historical sequences may uncover new amino
acid substitutions that are associated with antigenic change, such data are unlikely to improve the ability
of models to predict the antigenic phenotype of currently circulating viruses, which are evolving in new
ways, and also cannot be used to validate those predictions. Thus, unlike GoF approaches, alt-GoF
approaches are unable to substantially improve existing models by generating new experimental data
about relationships between antigenic phenotype and genotype. However, several completely different
types of data can increase the accuracy of these models and will complement improvements that can be
gleaned through the use of GoF data. These additional data sources include crystal structures for the HA
proteins from a wider variety of strains as well as data about how various amino acid substitutions affect
HA stability, which can be generated using in vitro, virus-free approaches.687
685 Hong M et al (2013) Antibody Recognition of the Pandemic H1N1 Influenza Virus Hemagglutinin Receptor Binding Site. J
Virol 87: 12471-12480

9.8.4.3 Benefits and limitations of alt-GoF approaches that can inform vaccine development
9.8.4.3.1 Vaccine development benefit 1: improve strain selection capabilities for seasonal influenza
vaccines
Alt-GoF approaches that have potential to benefit antigenic surveillance
GoF approaches have potential to benefit the strain selection process for seasonal influenza vaccines by
improving methods for predicting antigenic phenotype based on genotype, namely the use of molecular
markers of antigenic change and the use of computational models for sequence-based prediction of
antigenicity. As described above, alt-GoF approaches have limited abilities to improve either method,
relative to GoF approaches, though alt-GoF data complement GoF data to improve computational models.
Alt-GoF approaches that have potential to inform predictions of antigenic drift
GoF approaches can also benefit the strain selection process by improving methods for predicting
antigenic drift, which enables development of vaccines based on future, drifted strains, thereby increasing
the likelihood the vaccines match circulating strains at their time of deployment.Comparative sequence
analysis (alt-GoF) can also provide insight into antigenic evolution, which critically complements
laboratory evolution studies by generating insights that are directly relevant to the evolution of flu viruses
in human populations in nature. However, the ability of comparative sequence analysis to provide
mechanistic information about evolution is severely limited relative to GoF approaches. In addition,
analysis of wild type sequences cannot provide prospective information about the evolution of currently
circulating viruses. For both reasons, the use of comparative sequence analysis approaches is not
sufficient to improve the quality of existing models for antigenic evolution.
Alt-GoF approaches that have potential to improve strain selection capabilities through different
mechanisms
Alternative strategies for improving the quality of antigenic characterization data upon which strain
selection decisions are based are also being pursued. First, the Consortium for the Standardization of
Influenza Seroepidemiology (CONSISE) aims to standardize methods for the HAI assay, which would
ensure that antigenic data generated at disparate sites are more comparable.688 A second effort to improve
antigenic characterization data involves the development of alternative antigenic characterization assays,
which have greater potential for standardization and automation than the HAI assay; however, alternative
assays to date have had limited success.689
Several alternative approaches have potential to improve the strain selection process through completely
different mechanisms. First, increasing the timeliness, representativeness, and availability of surveillance
isolates would improve the accuracy of strain selection decisions by augmenting the quality of the input
data upon which those decisions are based. Key elements of efforts to strengthen influenza surveillance
systems include improving national surveillance systems, public health laboratories, and reporting and
virus sharing procedures in developing countries.690 To that end, between 2004 and 2014, the CDC
invested more than $150 million toward building sustainable lab capacity and NICs and other
688 Van Kerkhove MD et al (2013) The consortium for the standardization of influenza seroepidemiology (CONSISE): a global
partnership to standardize influenza seroepidemiology and develop influenza investigation protocols to inform public health
policy. Influenza Other Respir Viruses 7: 231-234
Switzerland, 1-3 April 2014. Vaccine 33: 4368-4382
690 Ibid.

international laboratories in over 40 less developed countries around the world.691 The WHO and other
WHO member countries also provide support, in the form of funding, technical expertise, and guidance.
However, given that resources for public health are limited and governments have many competing
priorities, sustaining and building upon gains in these areas that have occurred in the wake of the 2009
pandemic will continue to pose a major challenge.692,693
Other lines of research and new technologies have potential to fundamentally change current influenza
virological surveillance strategies and activities and may also lead to improved strain selection. For
example, an improved understanding of the spatiotemporal distribution of viruses and the factors that
influence the geographic spread of viruses could help target surveillance efforts and may also inform
prediction of whether and when antigenic variants detected in a particular region are likely to arise.694
Deep sequencing of surveillance isolates and systems biology approaches to analysis of such data may
provide insight into the role of host-pathogen interactions in the antigenic evolution of viruses, which
could also influence vaccination strategies and the strain selection process.695 In these and other cases,
because the state of the science and/or technology is preliminary, whether and when these approaches will
have a demonstrated impact on strain selection for seasonal influenza vaccines is unknown.
Alt-GoF approaches that have potential to improve the efficacy of seasonal flu vaccines through different
mechanisms
In addition to improving strain selection capabilities, several completely different strategies can be used
to increase the efficacy of seasonal flu vaccines. These strategies are described in detail in Sec. 9.5.4.2.2
and are briefly summarized here. First, a universal or broad-spectrum flu vaccine would obviate the need
for yearly production of strain-specific vaccines. However, influenza and vaccinology experts disagree
about the scientific feasibility of developing a universal vaccine, and one expert felt that a ten to twenty
year time frame for development is optimistic. Second, several scientific and technical advancements
could shorten production timelines for strain-specific vaccines, which would enable strain selection closer
to the start of flu season, presumably increasing the likelihood that the correct strains will be chosen. New
vaccine platforms, such as recombinant vaccines, can be rapidly scaled up and have shorter production
timelines than egg- and cell-based vaccines. However, the one recombinant vaccine on the market
accounts for less than 1% of total seasonal influenza vaccine produced annually, and although several
other virus-free vaccine platforms are in development, the length and expense of licensure processes for
new vaccines will delay their widespread availability. Incorporating adjuvants into existing egg- and cell-
based vaccines would allow for a smaller quantity of antigen to be used per vaccine dose, thus enabling
production of the same number of doses in a shorter period of time. However, no US-licensed seasonal
vaccines include adjuvants. Although an active area of research, adjuvanted vaccines must undergo
standard FDA licensing procedures for new vaccines and thus are unlikely to be broadly available in the
near future. Finally, GoF research that enhances virus production enables the development of higher-yield
CVVs, which shortens vaccine production timelines by increasing the rate of bulk antigen production.
694 Ibid.
695 Ibid.

9.8.4.3.2 Vaccine development benefit 2: Inform development of universal or broad-spectrum flu vaccines
GoF approaches to define the antigenic landscape of the HA protein may inform the development of
broad-spectrum or universal flu vaccines. Alternative approaches can also provide insight into which
regions of HA mutate to alter antigenicity and the spectrum of antigenic configurations the HA protein
can assume. First, attenuated reassortant strains (i.e. 6:2R strains with lab-adapted strains such as PR8)
can be used for forward genetic screens in lieu of wild type strains. As the antigenicity of 6:2R strains is
preserved relative to that of the parental seasonal flu strain, these strains are suitable for defining the
landscape of antigenic configurations that are possible for the HA protein; however, it is possible that
results may not translate to the wild type strain.
Alternative experimental approaches can also be used to study the antigenic landscape of the HA protein.
Comparative analysis of historical isolates can provide insight into mutations that are associated with
antigenic drift over time. However, this approach is constrained to studying the fraction of antigenic space
that the HA protein has explored in nature and cannot provide information about amino acid substitutions
that compromise virus viability, which is important knowledge for mapping the suite of substitutions that
are possible. Modeling approaches can, in principle, fully explore antigenic space but cannot yet
accurately predict antigenic phenotype from genotype nor the effects of HA mutations on protein stability
or viral fitness.
Completely different types of scientific data, generated through alt-GoF approaches, can also inform the
development of universal and broad-spectrum influenza vaccines. For example, one method for
identifying conserved epitopes involves identifying broadly neutralizing antibodies by characterizing the
ability of different monoclonal antibodies to neutralize a variety of strains, followed by antibody epitope
mapping.696 This knowledge can inform the development of multiple vaccine types. Another method
involves prediction of conserved immunogenic regions using in silico approaches, which has been used as
a basis for the development of peptide-based vaccines.697,698,699 Some of these vaccine candidates have
been shown to be immunogenic in animal studies and Phase I clinical trials.700,701,702 As all universal
vaccines are in early stages of development, whether these approaches will prove to be successful in
stimulating development of a safe, effective, and broad-spectrum influenza vaccine is unknown.
696 Zhu X et al (2013b) A unique and conserved neutralization epitope in H5N1 influenza viruses identified by an antibody
against the A/Goose/Guangdong/1/96 hemagglutinin. J Virol 87: 12619-12635
697 Gottlieb T, Ben-Yedidia T (2014) Epitope-based approaches to a universal influenza vaccine. Journal of autoimmunity 54:
15-20
698 Stoloff GA, Caparros-Wanderley W (2007) Synthetic multi-epitope peptides identified in silico induce protective immunity
against multiple influenza serotypes. European journal of immunology 37: 2441-2449
699 Adar Y et al (2009) A universal epitope-based influenza vaccine and its efficacy against H5N1. Vaccine 27: 2099-2107
700 Ibid.
701 Pleguezuelos O et al (2012) Synthetic Influenza vaccine (FLU-v) stimulates cell mediated immunity in a double-blind,
randomised, placebo-controlled Phase I trial. Ibid. 30: 4655-4660
702 Pleguezuelos O et al (2015) A Synthetic Influenza Virus Vaccine Induces a Cellular Immune Response That Correlates with
Reduction in Symptomatology and Virus Shedding in a Randomized Phase Ib Live-Virus Challenge in Humans. Clinical
and vaccine immunology : CVI 22: 828-835

approaches
immune pressure?
GoF approaches are uniquely capable of providing in-depth information about the evolutionary
mechanisms driving antigenic drift as well as prospective information about the evolution of currently
circulating viruses. In vivo approaches provide insight into antigenic drift in response to selective pressure
from the full complement of the immune system but may not translate to humans, while in vitro
approaches can provide information about antigenic changes that arise in response to selective pressure
from human antibodies but may not translate to complex, in vivo scenarios. In either case, lessons learned
in the laboratory may not translate to virus behavior in human populations in nature. In contrast,
comparative sequence analysis is uniquely capable of providing information about the antigenic evolution
of viruses in nature, but is constrained to reactively studying the evolution of historic viruses in limited
depth.
GoF approaches are uniquely capable of identifying amino acid substitutions that are necessary and
sufficient to alter antigenicity, which provides a critical foundation for the follow-up studies to elucidate
the biophysical basis of antigenic differences. Furthermore, GoF approaches represent the most efficient
and reliable method for uncovering mutations that cause antigenic drift in circulating strains and are
uniquely capable of exploring antigenic space to define which mutations do and do not lead to antigenic
changes, which can improve predictive modeling efforts. For the purpose of discovering mutations that
lead to antigenic change, GoF approaches can be conducted using attenuated 6:2R strains instead of
wildtype strains without compromising the quality and accuracy of the information that is generated.
Serial passaging of viruses in the presence of antibodies, a GoF approach, represents the only method for
mapping the antigenic sites of the HA protein in the context of a full virus. However, the fact that
mutations outside of antigenic sites may confer escape through long-range effects complicates
interpretation of mutational data from these experiments. In addition, the approach is relatively low-
throughput in that each passaging experiment enables identification of a single antigenic site. In contrast,
the use of cell surface display systems in yeast, bacteria, or phages represents a high-throughput method
for identifying the antigenic sites of particular mAbs or for comprehensively mapping the antigenic sites
on a given HA protein. Analysis of the crystal structures of HA-antibody complexes precisely reveals the
antibody binding site, but the resources needed and technical challenges associated with this approach
render it low-throughput. Additionally, the relevance of both in vitro approaches is limited by the fact that
that HA presentation may differ in the context of the full virus. Confirming the results of an in vitro
experiment requires determining whether mutating the proposed antigenic sites allows for escape from
antibody neutralization, a GoF approach.
GoF approaches that lead to evasion of existing natural or induced immunity have potential to benefit
surveillance of human seasonal influenza viruses in two ways: by increasing the utility of molecular
markers for antigenic drift and by improving the accuracy of existing models for predicting antigenic

phenotype from genotype. Attenuated reassortant strains (i.e. 6:2R strains with PR8) can be used in lieu
of wild type strains without diminishing these benefits.
GoF approaches are uniquely capable of discovering new amino acid substitutions that are necessary and
sufficient to alter antigenicity as well as determining whether markers are conserved in different strain
contexts, which collectively increase the predictive value of molecular markers for antigenic change.
Given the importance of genetic context for antigenic phenotype, whether such markers will ever be
strongly predictive is as-yet-unknown. Notably, GoF approaches to define the antigenic plasticity of the
HA protein are uniquely capable of addressing this question. Alternative experimental approaches cannot
provide causative data on molecular markers that contribute to altered antigenicity and are limited to
studying antigenic changes that have already occurred in nature, which significantly limits their utility for
this application.
GoF approaches are uniquely capable of generating experimental data about novel mutations that are
necessary and sufficient to confer antigenic change as well as validating predictions about antigenic
phenotype based on the sequences of currently circulating viruses, which will improve the accuracy of
existing predictive models. However, alternative types of data, including crystal structures of HA proteins
from additional strains, are also needed to improve the quality of existing models and will complement
gains achieved through the use of GoF approaches.
Together, molecular markers for antigenic change or predictive models can be used to supplement or
replace the use of phenotypic assays for characterizing the antigenicity of circulating seasonal influenza
viruses. Although molecular marker data currently informs the antigenic evaluation of surveillance
isolates, neither molecular markers nor computational models are robust enough to replace phenotypic
data (and may never be). However, use of these strategies to supplement phenotypic assays has potential
to improve the quantity and quality of antigenic characterization data that can be considered during
VCMs, which will increase the efficacy of seasonal flu vaccines as described below. Because molecular
marker data are currently used to aid interpretation of surveillance data, new data can be seamlessly
incorporated into the existing process, so that the only barrier to realization of this benefit is the need to
strengthen the state of the science. Influenza researchers involved in the strain selection process stated
that computational modeling could play an important role as well, once existing models are improved.703
Notably, GoF benefits to the quantity timeliness of antigenic characterization data considered during
VCM meetings rely on the generation of sequencing data at NICs (as opposed to WHOCCs). As less than
half of HA sequences for seasonal flu viruses are currently generated at NICs, full realization of this
benefit will necessitate further expansion of sequencing capabilities at NICs. 704
9.8.5.3 Benefits to vaccine development
9.8.5.3.1 Vaccine development benefit 1: Improve strain selection capabilities for seasonal influenza
vaccines
GoF approaches are uniquely capable of strengthening the predictive value of molecular markers for
antigenic change and play a critical role in improving models for predicting antigenic phenotype from
genotype as well as models for predicting antigenic drift. Although alternative experimental approaches
can provide other types of data that also strengthen predictive models, these data complement rather than
replace GoF data.
704 (2015w) Personal communication from WHOCC representative.

Advancing capabilities in these areas has the potential to benefit the strain selection process for seasonal
influenza vaccines in several ways. First, using sequence-based prediction of antigenic phenotype to
reinforce HAI assay results strengthens the robustness of antigenic characterization data, which provides a
stronger foundation for strain selection decisions. Second, given that genetic surveillance data are
increasingly available from NICs and other sample collection sites, shifting to sequence-based prediction
of antigenic phenotype in lieu of laboratory assays has potential to increase the timeliness and quantity of
surveillance data that are considered during VCMs. Third, predicting antigenic drift using models or
through experimental GoF approaches would enable the development of antigenically advanced vaccines
that are likely to match the circulating strains when vaccines are deployed, thereby increasing vaccine
efficacy. However, full realization of these benefits necessitates further expanding sequencing capabilities
at NICs.
Current experimental and modeling efforts cannot yet predict antigenic phenotype from genotype or the
timing and direction of antigenic drift. Whether and when such capabilities will be sufficiently accurate to
be incorporated into the strain selection process is unknown and depends both on scientific advancements
and inherent features of influenza biology. Namely, the antigenic plasticity of the HA protein is not well-
characterized but governs the feasibility of each of these predictive efforts. Notably, GoF efforts are also
essential for advancing understanding of the antigenic landscape of HA.
Several alternative approaches have potential to improve the strain selection process through different
mechanisms. First, efforts to standardize the HAI assay and to develop new antigenic characterization
assays are ongoing, both of which have potential to improve the quality of antigenic characterization data.
However, these alternative assays are not yet viable replacements for the HAI assay, and the degree to
which increased standardization of the HAI assay will improve data quality is uncertain. Initiatives to
strengthen global influenza surveillance systems have potential to improve the timeliness,
representativeness, and quantity of surveillance isolates that can be considered at VCMs but face
considerable funding and political barriers. Finally, new technologies such as deep sequencing have the
potential to revolutionize influenza virological surveillance activities and may improve strain selection
capabilities through unexpected mechanisms. Each of these alternative approaches either complements
GoF approaches or addresses different shortcomings in the strain selection process.
Given the complexities involved in coordinating global influenza surveillance and making strain selection
decisions under the time pressures imposed by vaccine production timelines, as well as the significant
uncertainties in whether and when both GoF and alt-GoF approaches will yield demonstrable benefits to
the process, pursuing both GoF and alt-GoF strategies in tandem will ensure that strain selection
capabilities are advanced rapidly and to the greatest extent possible.
Finally, several alternative approaches have potential to improve the efficacy of seasonal influenza
vaccines through completely different mechanisms. Universal vaccines represent the only strategy with
potential to fully “solve” the vaccine mismatch problem but are in early stages of development and
represent a long-term solution at best. Several approaches, namely the development of virus-free
vaccines, the incorporation of adjuvants into existing vaccines, and the development of higher-yield
vaccine viruses through GoF approaches that enhance virus production, have potential to shorten
production timelines for strain-specific vaccines. This adjustment to manufacturing schedules could
enable strain selection closer to the start of flu season, which presumably will increase the likelihood of
vaccine match. Importantly, all of these approaches complement efforts to improve strain selection
capabilities because each approach addresses different underlying gaps in current scientific and technical
capabilities that contribute to vaccine mismatch. Thus, influenza vaccine experts recommend pursuing all
of these approaches as part of comprehensive strategy for improving the quality of seasonal influenza
vaccines.

9.8.5.3.2 Vaccine development benefit 2: Inform development of universal or broad-spectrum flu vaccines
GoF approaches are uniquely capable of defining the antigenic landscape of the influenza HA protein –
that is, the spectrum of antigenic configurations that HA can assume and which regions of HA are capable
of mutating while preserving virus viability. These data may inform the development of broad-spectrum
influenza vaccines, which protect against a large fraction of the possible antigenic space, or universal
influenza vaccines, which target regions of the protein that are unable to mutate and thus are drift-
resistant. Alternative experimental approaches have significant limitations. Attenuated reassortant strains
can be used to explore possible antigenic configurations, but results regarding the fitness consequences of
mutations may not translate to wild type strains. Comparative analysis of historical isolates is limited to
the fraction of antigenic space that has been explored in nature and cannot provide information on
mutations that compromise virus viability. While virus-free approaches can be used to explore new
antigenic space, these approaches do not reveal the fitness consequences of mutations either. Finally,
existing models cannot accurately predict antigenic phenotype from genotype or predict the fitness
consequences of particular mutations.
Mapping the antigenic landscape of the HA protein represents a labor-intensive project, and whether
vaccine development strategies based on the information gleaned from this approach will be successful is
unknown. Other strategies for developing broad-spectrum and universal vaccines, such as in silico
prediction of conserved epitopes for the development of peptide-based vaccines, have shown promise. All
universal/broad-spectrum vaccine candidates are in early stages of development, and which strategy is
likely to be most successful is unknown. Given the challenges for developing universal/broad-spectrum
vaccines, pursuing all experimental approaches that support vaccine development in tandem, including
GoF approaches, will maximize the likelihood of success, which could have large public health impacts.
9.9 Influenza viruses: Benefits of GoF Research that Leads to Evasion of Vaccines
9.9.1 Summary
influenza vaccines in development. Such GoF studies were found to have unique benefits to the
development of new influenza vaccines. No alternative approaches were identified that can provide
similar benefits. Chapter 9.9 provides an overview of these benefits, including basic background and
supporting information; a fully referenced and more thorough discussion of these benefits can be found in
Appendix IV Sec. 15.6.
9.9.1.1 Benefits of GoF that leads to evasion of vaccines to vaccine development
• GoF approaches are uniquely capable of determining whether and how readily influenza viruses
can acquire mutations to escape neutralization by candidate broad-spectrum or universal
influenza vaccines, a critical aspect of testing the potential field efficacy of vaccine candidates.
• There are no alt-GoF approaches that can provide similar information.
9.9.2 Overview of GoF Research Landscape: Evasion of Vaccines
9.9.2.1 Serial passaging of viruses in the presence of post-vaccination sera
Serial passaging of a virus in cells in the presence of animal sera produced in response to a vaccine or in
vaccinated animals may lead to the emergence of viruses that are resistant to neutralization by vaccine-

induced antibodies. This approach is used to test whether and how readily viruses can evolve to evade
vaccines in development, for example new vaccine platforms that are more broad-spectrum or resistant to
drift than current influenza vaccine platforms, which is an important indicator of the potential field
efficacy of the vaccine.
In this section, the potential benefits of GoF research that leads to evasion of vaccines in each benefit
category listed in the NSABB Framework are discussed.
This GoF approach is solely focused on understanding how a virus evolves in response to immune
pressure from a vaccine under development. As a result, insights gleaned from this approach do not
benefit scientific knowledge, surveillance or policy decisions (because the vaccine has not yet been
deployed) or the development of therapeutics and diagnostics.
9.9.3.1 Benefits and Limitations of GoF Approaches to Vaccine Development
GoF approaches that lead to evasion of vaccines in development benefit the development of new
influenza vaccines. Specifically, these approaches demonstrate whether and how readily viruses can drift
to escape neutralization by new vaccine candidates, which is an important indicator of their potential field
efficacy relative to existing vaccines.
9.9.3.1.1 Shortcomings in existing influenza vaccines
Because existing influenza vaccines are strain-specific, new seasonal flu vaccines must be produced
annually in order to accommodate antigenic drift of circulating influenza viruses, and new pandemic flu
vaccines must be produced in response to the emergence of a novel pandemic strain. The long production
timelines for existing influenza vaccines critically limit the mitigating impact of influenza vaccination on
the morbidity and mortality associated with influenza outbreaks, as discussed in Sec. 9.5.3.3.1. For these
reasons, the influenza research and public health communities are strongly interested in developing a
broad-spectrum or universal flu vaccine.705,706 Demonstrating whether such vaccine candidates are more
resistant to antigenic drift than existing vaccines is a critical aspect of testing the potential field efficacy
of these vaccine candidates.707
9.9.3.1.2 Benefits and limitations of GoF approaches
Serial passaging of viruses in cells, in the presence of sera from vaccinated animals, or in vaccinated
animals may lead to the emergence of mutant viruses that can no longer be neutralized by vaccine-
induced antibodies. Sequencing of emergent escape mutants provides insight into how readily viruses can
acquire mutations that confer escape from protective vaccination (i.e. how many mutations are needed to
705 Rudolph W, Ben Yedidia T (2011) A universal influenza vaccine: where are we in the pursuit of this "Holy Grail"? Human
vaccines 7: 10-11
707 Ibid.

escape neutralization). Follow-up studies characterizing other properties of emergent escape viruses
relative to the parental virus, such as fitness, may provide additional insight into how likely vaccine
escape mutants are to emerge and persist in human populations. In vitro studies provide a proof of
principle demonstration of whether viruses can mutate to escape vaccines, but virus behavior in response
to relatively simple selection pressures may not translate to human populations. In vivo studies involve
complex selection pressures that more closely mimic those that a virus will encounter during infection of
a vaccinated human host, but results in representative animal models may not translate to human disease.
9.9.3.2 Economic benefits of GoF approaches
GoF benefits to the development of new vaccines may have downstream economic benefits. Economic
benefits were not explicitly evaluated in this report.
No alternative approaches are capable of evaluating whether viruses can acquire mutations to escape
neutralization by candidate vaccines prior to field deployment of the vaccine.
approaches
Taken together, GoF approaches are uniquely capable of determining whether and how readily influenza
viruses can acquire mutations to escape neutralization by candidate broad-spectrum or universal influenza
vaccines, a critical aspect of testing the potential field efficacy of vaccine candidates.
9.10 Influenza Viruses: Benefits of GoF Research that leads to Evasion of Therapeutics
9.10.1 Summary
therapeutics, including licensed therapeutics and therapeutics in development. Such GoF studies were
found to generate scientific knowledge; to inform surveillance of circulating seasonal and animal
influenza viruses, which guides therapeutic recommendations for seasonal flu and decision-making about
pandemic preparedness initiatives, respectively; to benefit the production of influenza vaccines; and to
benefit the development of new therapeutics. Alt-GoF approaches that may generate similar benefits were
also identified and analyzed. At present, GoF studies resulting in evasion of existing natural or induced
adaptive immunity have unique benefits to scientific knowledge, surveillance, and therapeutic
development, though full realization of GoF benefits to surveillance requires scientific advancements and
expansion of global public health surveillance networks. Chapter 9.10 provides an overview of these
benefits, including basic background and supporting information; a fully referenced and more thorough
discussion of these benefits can be found in Appendix IV Sec. 15.7.
9.10.1.1 Benefits of GoF Research that Leads to Evasion of Therapeutics to Scientific Knowledge
• GoF approaches:
o Are the most efficient and effective strategies for discovering novel mutations that confer
resistance to antivirals.

o Are uniquely capable of identifying mutations that are necessary and sufficient to confer
antiviral resistance across multiple strain contexts, which provides a strong foundation for
follow-up studies to elucidate the mechanisms underlying antiviral resistance.
o Attenuated reassortant strains may be used in lieu of wild type strains for many experiments
investigating the mechanistic basis of resistance, but results may not be recapitulated in the
context of the wild type viruses.
o Are the most efficient and effective strategy for gaining in-depth insight into the viral and
host selection pressures that shape the emergence and spread of antiviral resistance.
o Are capable of discovering novel mutations associated with antiviral resistance but have
limitations relative to GoF approaches.
o Equally capable of establishing a causal link between a particular mutation and antiviral
resistance, but are limited in their ability to demonstrate that the function of markers is
conserved across strain contexts by the breadth of antiviral resistant strains that exist in
nature.
o Comparative analysis of patient isolates over the course of antiviral treatment can provide in-
depth insight into the evolution of antiviral resistance, but such studies are relatively rare and
may not translate to the general population.
o Other alt-GoF approaches provide limited mechanistic insight about the evolutionary
pressures driving emergence of antiviral resistance.
9.10.1.2 Benefits of GoF Research that Leads to Evasion of Therapeutics to Surveillance
• GoF approaches:
o Provide unique benefits for strengthening the predictive value of molecular markers for
antiviral resistance, thereby improving their utility for interpreting surveillance data.
o Molecular markers (discovered and validated through GoF approaches) have potential to
strengthen the quality and timeliness of antiviral resistance information about viruses
collected through surveillance, by:
 Corroborating phenotypic characterization data, and
 Enabling sequence-based prediction of the antiviral resistance phenotype, prior to the
availability of phenotypic data.
o Full realization of the benefits of GoF approaches to surveillance is subject to expansion of

sequencing capabilities at public health laboratories that collect clinical specimens.
o Are significantly limited in their ability to strengthen the predictive value of molecular
markers for antiviral resistance.
o Phenotypic assays play a critical role in evaluating the antiviral resistance of surveillance
isolates because they provide direct information about the degree of antiviral resistance of a
particular strain, so that sequence-based predictions should be confirmed whenever possible.

9.10.1.3 Benefits of GoF Research that Leads to Evasion of Therapeutics to Decision-Making in
Public Health Policy
• GoF approaches:
o Data on the prevalence of antiviral-resistant seasonal strains, collected through surveillance,
informs therapeutic recommendations developed by the CDC. Both molecular marker data
(GoF) and phenotypic data (alt-GoF) inform interpretation of surveillance data.
o Improving the practice of using molecular marker (GoF) may enable a larger quantity of
strains to be assessed for their antiviral sensitivity, which would improve the ability to detect
and track the emergence of rare antiviral-resistant strains.
o The observation of antiviral resistance in an animal influenza strain, coupled to other factors
indicative of increased pandemic potential, may trigger downstream responses such as
applying for Emergency Use Authorization (EUA) for antivirals in development.
 Using molecular markers for antiviral resistance (GoF) enables a rapid risk assessment
based on sequence data when a novel virus first emerges in human populations, which
can provide a several week head start on downstream response activities.
o Results from phenotypic testing of seasonal flu surveillance isolates critically informs
therapeutic guidelines for seasonal flu. A subset of surveillance isolates are subjected to
phenotypic testing for antiviral resistance.
o Confirming whether an animal influenza strain is antiviral-resistant through phenotypic

testing is critical for pandemic risk assessments and downstream decision-making, but results
from phenotypic assays may be delayed relative to the publication of sequence data due to
delays in shipping of virus samples.
9.10.1.4 Benefits of GoF Research that Leads to Evasion of Therapeutics to Vaccine Development
• GoF approaches:
o Are the most efficient and effective way to discover novel markers for antiviral resistance and
can establish a causal link between a particular mutation and antiviral resistance across many
strain contexts. These conserved markers can be mutated out of vaccine viruses to increase
the safety of the vaccine production process.
o Can be used to establish a causal link between a particular trait and antiviral resistance, but
their ability to demonstrate that a particular marker is conserved across strain contexts is
limited by the breadth of antiviral resistant strains that exist in nature.
9.10.1.5 Benefits of GoF Research that Leads to Evasion of Therapeutics to Therapeutic Development
• GoF approaches:
o Are uniquely capable of screening potential therapeutic candidates based on how readily
antiviral resistance emerges, which is an important indicator of the potential field efficacy of
a therapeutic.

o Are uniquely capable of determining whether the acquisition of resistance to a therapeutic
candidate increases the infectivity, transmissibility, or virulence of a virus, which is an
important aspect of safety testing of the therapeutic candidate.
o Are uniquely capable of determining the genetic threshold for resistance to a new therapeutic,
prior to field deployment of that therapeutic.
o Are uniquely capable of identifying the viral target of a novel therapeutic with an unknown
mechanism of action, which is valuable for determining the mechanism of action of
therapeutic candidates identified through unbiased high-throughput screens.
o Are uniquely capable of determining the therapeutic dose that is least likely to lead to the
acquisition of antiviral resistance as well as determining whether combination therapies better
prevent the emergence of resistant viruses than individual therapies, which informs the
development of therapeutic strategies that will be effective for a longer time in the field.
o X-ray crystallography and photoaffinity crosslinking are limited to the study of therapeutics
with known viral targets, and inferring mechanistic information based on static data about
drug-viral interactions may be difficult.
o RNAi screens to identify host factors that are required for the antiviral activity of a
therapeutic provide indirect information about the mechanisms of therapeutics that target
viral proteins.
9.10.2 Overview of GoF research landscape: evasion of therapeutics
9.10.2.1 Serial Passaging of Viruses in the Presence of Therapeutics
Serial passaging of viruses in the presence of a therapeutic may lead to the acquisition of mutations that
allow the virus to evade inhibition by the therapeutic. This approach is performed to determine whether
and how readily a virus evolves resistance in response to selective pressure from a therapeutic and to
identify mutations that confer resistance, which provides a foundation for follow-up studies investigating
the mechanism of action of the therapeutic and the mechanistic basis of antiviral resistance. When
passaging experiments are performed using a new therapeutic candidate with an unknown viral target, this
information also helps to identify the therapeutic target, as resistance mutations are most likely to arise in
the target protein. Serial passaging approaches have been performed using cell culture, animal models,
and (rarely) human challenge experiments.
9.10.2.2 Targeted Modification of Viruses to Introduce Mutations that are Expected to Lead to Evasion
of Therapeutics
A second approach involves targeted genetic modification of a virus to introduce mutations that are
associated with antiviral resistance, which may have been identified through GoF approaches such as
serial passaging or through alt-GoF approaches such as comparative analysis of sequences from patients
who did and did not respond to antiviral treatment. This experiment serves to demonstrate that a particular
mutation or set of mutations is necessary and sufficient to enhance antiviral resistance, which provides a
foundation for follow-up studies investigating the mechanistic basis of antiviral resistance.

9.10.3.1 Benefits and Limitations of GoF Approaches to Scientific Knowledge
Only one class of licensed antivirals are recommended for therapeutic use against seasonal influenza
viruses: the neuraminidase inhibitors (NAIs), which inhibit the activity of the NA protein.708,709,710
Although most circulating strains have been sensitive to all three licensed NAIs during recent flu seasons,
strains that are resistant to one or more NAIs have been observed in nature in A/H1N1,711 A/H3N2,712 and
B strains.713 Resistance has been linked to a variety of mutations, and in most cases, the mechanisms
underlying drug resistance are not well understood. In addition, the factors that shape whether resistant
strains will emerge, spread and persist in human populations, including the contribution of viral factors
such as the relative fitness of resistant strains, are unknown.
In this section, the ability of GoF approaches, versus alternative experimental approaches, to address two
unanswered questions in this field are addressed:
• What are the genetic traits underlying resistance to NAIs, and what is the mechanistic basis of
resistance?
• What selection pressures shape whether and how readily antiviral-resistant strains arise and
spread in nature?
9.10.3.1.1 Scientific Knowledge Gap #1: What are the Genetic Traits Underlying Resistance to NAIs, and
What is the Mechanistic Basis of Resistance?
Serial passaging of viruses in the presence of one or multiple therapeutics may lead to the emergence of
viruses that are resistant to inhibition by the therapeutic. Sequencing emergent antiviral-resistant viruses
enables the identification of novel mutations that are sufficient to confer resistance. Selection for
resistance studies can be carried out in vitro, in vivo, in animals or through human challenge experiments.
(Human challenge experiments are rare and have only been conducted using human seasonal strains.)
Notably, in vitro and in vivo selection approaches equally enable the identification of mutations
associated with antiviral resistance, though the in vitro approach is faster and cheaper. The in vitro
approach is highly efficient and can be carried out using any virus strain, including currently circulating
strains. Importantly, as multiple mutations may arise during passaging, follow-up studies may be needed
Forward genetic screens, which involve random mutagenesis of the NA genes from antiviral-sensitive
strains followed by screening of mutants to identify those with reduced antiviral susceptibility, represent
another GoF approach for discovering novel mutations that confer antiviral resistance. The screening

709 Kim CU et al (1997) Influenza neuraminidase inhibitors possessing a novel hydrophobic interaction in the enzyme active
site: design, synthesis, and structural analysis of carbocyclic sialic acid analogues with potent anti-influenza activity. J Am
Chem Soc 119: 681-690
710 Li W et al (1998) Identification of GS 4104 as an orally bioavailable prodrug of the influenza virus neuraminidase inhibitor
GS 4071. Antimicrob Agents Chemother 42: 647-653
711 Gubareva LV et al (2001) Selection of influenza virus mutants in experimentally infected volunteers treated with
oseltamivir. J Infect Dis 183: 523-531
712 Abed Y et al (2006) Impact of neuraminidase mutations conferring influenza resistance to neuraminidase inhibitors in the
N1 and N2 genetic backgrounds. Antiviral therapy 11: 971-976
713 Fujisaki S et al (2012) A single E105K mutation far from the active site of influenza B virus neuraminidase contributes to
reduced susceptibility to multiple neuraminidase-inhibitor drugs. Biochem Biophys Res Commun 429: 51-56

approach is less efficient that the selection approach but may enable the discovery of rare antiviral
resistance mutations that might be out-competed during a selection experiment due to fitness defects.
Depending on the mutagenesis strategy used, follow-up studies may be needed to determine which
mutation(s) are responsible for the antiviral resistance phenotype. Additionally, for both the serial
passaging and forward genetic screen approaches, results may not translate to other strain contexts.
Targeted genetic modification of parental viruses to introduce mutations associated with antiviral
resistance, followed by phenotypic characterization of the antiviral sensitivity of mutant viruses, is used to
demonstrate that a mutation or set of mutations is necessary and sufficient to confer resistance.
Subsequently, to determine whether the phenotypic consequences of mutations are functionally
generalizable across multiple virus strains, targeted mutagenesis can be used to introduce mutations into
new virus strains, followed by characterization of antiviral sensitivity. Together, these results provide a
strong foundation for follow-up biochemical, cell biological, structural, and other studies to determine the
mechanistic basis of antiviral resistance.
9.10.3.1.2 Scientific Knowledge Gap #2: What Selection Pressures Shape Whether and How Readily
Antiviral-Resistant Strains Arise and Spread in Nature?
Serial passaging of viruses in the presence of one or more therapeutics to select for antiviral-resistant
strains provides insight into whether and how readily antiviral resistance arises. Follow-up experiments to
characterize other phenotypic properties of emergent resistant viruses, such as fitness, virulence, and
transmissibility, may provide insight into how likely resistant viruses are to emerge, spread, and persist in
human populations. These experiments have been conducted in vitro and in vivo, through animal
experiments and human challenge experiments. Due to the simple selection pressures encountered by
viruses during passage in cell culture, the in vitro approach is less useful than the in vivo approach for
understanding how selection pressures in human populations are likely to drive the emergence and spread
of antiviral-resistant viruses. The ability to gain direct insight into emergence of resistance in humans
through human challenge experiments is valuable, but ethical considerations severely constrain the
number and scope of experiments that can be carried out. Animal experiments provide a controlled
system for studying the emergence of resistance under complex selection pressures, including identifying
resistance mutations that arise but are negatively selected within or between hosts. However, results in
animal models may not translate to human populations.
9.10.3.2 Benefits and Limitations of GoF Approaches to Surveillance
Through influenza surveillance, public health professionals monitor the appearance and prevalence of
NAI-resistant strains of seasonal influenza viruses circulating in global populations and of animal
influenza viruses that have caused human infections. Data on the prevalence of antiviral-resistant seasonal
strains informs therapeutic recommendations developed by the CDC (i.e. which of the three FDA-
approved NAIs should be recommended for treatment). 714 In the context of surveillance for zoonotic
influenza infections in humans, data on antiviral resistance informs decision-making about pandemic
preparedness initiatives because antiviral resistance is one of the risk elements considered in a pandemic
risk assessment.
Resistance to NAIs can be assessed in two ways: through laboratory testing of NAI resistance using the
fluorometric 20-(4-methylumbelliferyl)-a-D-N-acetylneuraminic acid (MUNANA) assay or other assays
and/or by inspecting sequences for the presence of mutations that are known to confer NAI resistance
(“molecular markers”). This section evaluates the potential for GoF approaches to improve the practice of
714 Centers for Disease Control and Prevention. Influenza Antiviral Medications: Summary for Clinicians.
http://www.cdc.gov/flu/professionals/antivirals/summary-clinicians.htm. Last Accessed: November 4, 2015

using molecular markers for antiviral resistance and the relative utility of sequence-based predictions
versus phenotypic assays for surveillance for antiviral resistant viruses. The public health actions that are
taken downstream of this surveillance are described in the “informing policy decisions” section, below.
9.10.3.2.1 Benefits of using molecular markers for antiviral resistance to infer antiviral resistance
phenotype from genotype
The practice of using molecular markers for antiviral resistance to predict the antiviral resistance
phenotype of viruses detected through surveillance provides several advantages relative to the use of
phenotypic assays. In particular, the ability to sequence clinical samples is valuable because the
composition of viral quasispecies changes during the virus isolation process, which can mask the presence
of antiviral resistant strains in mixed infections if those resistant strains are lost during the virus isolation
process.715 In addition, the expansion of viral sequencing capabilities at the “base” level of the
surveillance system (NICs and other diagnostic labs) means that sequencing data are increasingly
available prior to phenotypic data, which are generated at WHOCCs.716 Thus, the use of molecular
markers can enable a more timely assessment of the antiviral resistance phenotype of strains detected
through surveillance. Finally, as sequencing becomes cheaper and easier, the number of surveillance
isolates that are sequenced is likely to increase, such that using molecular markers will enable
consideration of a larger number of viruses than can be subjected to phenotypic characterization assays.717
9.10.3.2.2 Current utility and shortcomings of using molecular markers for antiviral resistance
NAI resistance can arise from one or two mutations, and many mutations have been identified that confer
resistance to one or multiple NAIs. Several markers for NAI resistance have been shown to be
functionally generalizable, conferring resistance in multiple strain contexts.718 In the experience of
influenza researchers and government officials involved in surveillance, the presence of such a validated
antiviral resistance marker is strongly predictive antiviral resistance. However, the absence of a known
resistance marker is not necessarily predictive of antiviral sensitivity, as it is likely that additional
mutations can lead to resistance. This lack of knowledge about the mutational landscape that permits
evolution of antiviral resistance limits the current utility of sequence-based approaches for predicting
resistance. Moreover, validating known markers in additional strain contexts will further strengthen their
predictive value. As discussed in detail in Sec. 9.10.3.1.1, above, both GoF and alt-GoF approaches can
provide insight into these scientific questions. The relevant findings are summarized here.
9.10.3.2.3 Potential Benefits of GoF approaches to the practice of using molecular markers for antiviral
resistance
GoF approaches represent the most efficient and effective strategy for discovering novel mutations that
give rise to antiviral resistance and are uniquely capable of confirming that particular mutations are
necessary and sufficient to confer resistance in multiple strain contexts. Notably, for mutations that confer
resistance by altering the function of the NA protein (i.e. versus altering NA expression levels or through
epistatic effects), these experiments can be performed using attenuated reassortant strains, though results
may not be recapitulated in the context of the wild type strain. Taken together, these approaches
strengthen the predictive value of molecular markers for antiviral resistance, thereby improving their
utility for interpreting surveillance data.

716 Ibid.
718 Boivin G (2013) Detection and management of antiviral resistance for influenza viruses. Influenza and Other Respiratory
Viruses 7: 18-23

9.10.3.3 Benefits and Limitations of GoF Approaches to Decision-Making in Public Health Policy
GoF approaches have potential to benefit surveillance for antiviral resistant strains by improving the
practice of using molecular markers for antiviral resistance to infer antiviral resistance from genotype.
Surveillance for antiviral resistant strains informs downstream decision-making related to public health
practice and policy. Namely, data on the prevalence of antiviral-resistant seasonal strains informs
therapeutic recommendations developed by the CDC, and antiviral resistance is one of the risk elements
considered in a pandemic risk assessment of an animal influenza virus. This section describes each of
these applications, which illustrate the ultimate public health impacts associated with GoF benefits to
surveillance.
9.10.3.3.1 Benefits to Decision-Making Related to Seasonal Flu Strains
The CDC monitors the prevalence of antiviral resistance among circulating strains to inform their
recommendations to clinicians for the use of influenza antivirals. Although recent seasonal outbreak
strains have remained susceptible to all three NAIs, sporadic cases of antiviral resistant viruses continue
to be detected. 719 Current antiviral treatment guidelines do not recommend particular NAIs; however, an
increase in the prevalence of singly-resistant strains could trigger a recommendation change. As antivirals
are most effective when given within 48 hours of symptom onset, the CDC recommends initiating
antiviral treatment prior to laboratory confirmation of influenza (i.e. without knowledge of antiviral
susceptibility).720 Given that, antiviral treatment recommendations based on reliable knowledge about the
prevalence of resistance to particular antivirals among circulating strains is essential for the success of
therapeutic treatment. Currently, a subset of the influenza viruses that are collected by WHOCCs are sent
to CDC for antiviral susceptibility testing.721 As discussed above, phenotypic assay results are often
corroborated by inspection of sequences for the presence of molecular markers associated with antiviral
resistance, and the use of molecular markers may expand the number of viruses that can be evaluated for
their antiviral susceptibility as the number of surveillance isolates that are sequenced increases. This
increase will provide a stronger foundation for antiviral treatment recommendations and may enhance
detection of rare antiviral-resistant variants to increase awareness.
9.10.3.3.2 Benefits to Decision-Making Related to Pandemic Influenza
Antiviral resistance is one of the risk elements considered in pandemic risk assessments of animal
influenza viruses, which inform downstream decision-making about investments in pre-pandemic
preparedness initiatives (discussed in detail in Sec. 9.6.3.3.2 and Sec. 9.6.3.3.3). The antiviral resistance
risk element does not contribute to the likelihood that an animal virus will emerge to efficiently infect and
transmit in humans and moderately contributes to the assessment of the expected consequences of an
emergence event. For example, in a recent risk assessment of avian H7N9, avian H1N1, and swine
H3N2v viruses, the antiviral resistance element was worth approximately one-third as much as the most
highly weighted disease severity element.722 Stakeholders involved in the pandemic risk assessment
process emphasized that antiviral resistance does not increase risk a priori but rather is important when
coupled to other factors indicative of increased pandemic potential, such as a high number of human
infections or enhanced transmissibility or virulence in ferrets. In this case, the observation of antiviral
resistance may trigger USG representatives to initiate the process of applying for Emergency Use
720 Ibid.
721 Centers for Disease Control and Prevention. Antiviral Drug Resistance among Influenza Viruses.
http://www.cdc.gov/flu/professionals/antivirals/antiviral-drug-resistance.htm. Last Accessed: November 4, 2015

Authorization (EUA) from the FDA for antivirals in development, to ensure that effective antivirals will
be available if the strain under evaluation spreads to cause a pandemic. When evaluating antiviral
resistance, stakeholders consider both phenotypic and genetic data when possible but noted that the ability
to conduct a rapid risk assessment using molecular markers is valuable when strains first emerge and
sequences are published prior to the receipt of viral isolates. This rapid assessment can provide a several
week head start on the EUA process. As EUAs may be issued within days following a request if the FDA
has worked with government partners on a “pre-EUA” package, this head start could significantly impact
the timing of availability of antivirals in the event of a pandemic.723,724
9.10.3.4 Benefits and Limitations of GoF Approaches to the Development of Vaccines
9.10.3.4.1 Vaccine Development Benefit 1: Targeted Mutagenesis to Remove Antiviral Resistance

Markers from Vaccine Viruses
Vaccine viruses comprise the HA and NA genes from the wild type strain of interest and the remaining
six genes from a vaccine backbone virus such as PR8. Mutations that confer resistance to NAIs, the one
approved class of influenza antivirals that are recommended for use in the U.S., arise in the NA
gene.725,726,727,728 If the wild type NA gene contains conserved markers for NAI resistance, these markers
can be removed through targeted deletion or mutagenesis to increase the safety of the vaccine production
process. (Of note, most influenza vaccines produced in the U.S. are inactivated, thus whether a vaccine
strain is sensitive or resistant to antivirals has no impact on the safety of the vaccine itself.)
GoF approaches represent efficient and effective strategies for the discovery of new antiviral resistance
markers but may uncover mutations that do not yet exist in nature, which is not relevant for this
application because vaccine viruses are based on wild type viruses. Subsequently, targeted mutagenesis of
antiviral-sensitive strains to introduce mutations expected to confer antiviral resistance, can be used to
demonstrate that a mutation or set of mutations is necessary and sufficient to confer antiviral resistance.
The establishment of such a causal link is critical for the application of this information to vaccine
development.
9.10.3.5 Benefits and Limitations of GoF Approaches to the Development of Therapeutics
Given that only one class of antivirals is licensed and recommended for use in the U.S. and strains that are
resistant to one or more therapeutics in this class have been detected in nature, there is an urgent need for
the development of new therapeutics against influenza viruses.729
GoF approaches that lead to evasion of therapeutics have the potential to benefit the development of new
therapeutics in several ways:
723 Administration FaD. Guidance - Emergency Use Authorization of Medical Products.

http://www.fda.gov/RegulatoryInformation/Guidances/ucm125127.htm. Last Update Accessed November 10, 2015.
724 (2015y) Personal communication from FDA representative.
725 Baz M et al (2010) Effect of the neuraminidase mutation H274Y conferring resistance to oseltamivir on the replicative
capacity and virulence of old and recent human influenza A(H1N1) viruses. J Infect Dis 201: 740-745
726 Kaminski MM et al (2013) Pandemic 2009 H1N1 influenza A virus carrying a Q136K mutation in the neuraminidase gene
is resistant to zanamivir but exhibits reduced fitness in the guinea pig transmission model. Journal of virology 87: 1912-
1915
727 Baz M et al (2006) Characterization of Multidrug-Resistant Influenza A/H3N2 Viruses Shed during 1 Year by an
Immunocompromised Child. Clin Infect Dis 43: 1555-1561
728 Hai R et al (2013) Influenza A(H7N9) virus gains neuraminidase inhibitor resistance without loss of in vivo virulence or
transmissibility. Nat Commun 4

• GoF approaches can be used to screen therapeutic candidates based on how readily various
candidates acquire resistance and provide information about whether the emergence of resistance
enhances the transmissibility or virulence of resistant viruses, an important aspect of safety
testing.
• GoF approaches provide information about the potential field efficacy of the therapeutic and the
mechanism of activity of the therapeutic, both of which are critical components of an
Investigational New Drug application to the FDA.
• GoF approaches can provide insight into the therapeutic dosing regimens and combination
therapies (e.g. cocktails of monoclonal antibodies) that are the least likely to permit evolution of
resistance.
9.10.3.5.1 Therapeutic Development Benefit #1: Inform Development of Therapeutic Candidates
Given the high mutation rate of influenza viruses, viruses can readily acquire mutations to many
therapeutics. Screening potential therapeutics based on how readily antiviral resistance emerges provides
one mechanism for differentiating between therapeutic candidates based on their likely field efficacy.
Prior to field deployment of a therapeutic, serially passaging viruses in the presence of therapeutic, a GoF
approach, is uniquely capable of determining whether and how readily resistance arises. Furthermore, as
resistance is expected to arise in human populations following deployment of the therapeutic, determining
whether resistance enhances the infectivity, transmissibility, or virulence of a virus is an important aspect
of safety testing of the therapeutic candidate.
9.10.3.5.2 Therapeutic Development Benefit #2: Facilitate Regulatory Approval of Therapeutic

Candidates
investigational product, and the genotypic and phenotypic characterization of resistant viruses.730 GoF
approaches have the potential to support two aspects of an IND application for therapeutics in
development: (1) determination of the mechanism of action of a therapeutic, and (2) the in vitro selection
of resistant viruses.
Determining the Mechanism of Action of a Therapeutic
The FDA recommends that a drug’s mechanism of action be “well-characterized” prior to the start of
Phase I clinical trials and requests this information as a component of an IND application, the first step of
the licensing process.731 The influenza field is pursuing multiple strategies for developing new therapeutic
candidates, including the deliberate design or selection of therapeutics targeting specific viral or host
proteins and high-throughput screening of libraries of small molecules to identify compounds that reduce
viral replication in vitro. In the former case, the drug target of the therapeutic candidate is known, while
in the latter case, the therapeutic target is unknown. GoF approaches can be used to gain insight into the
731 Ibid.

mechanism of activity of therapeutics that directly target virus proteins, thus benefitting the development
of new drugs.
Passaging viruses in cells in the presence of a therapeutic is a classic method for generating viruses that
can evade the inhibitory action of the therapeutic, thus constituting a GoF approach. Viruses are then
sequenced to identify mutations that arose, and if multiple mutations are present, mutations are re-
introduced into the parental strain individually and in combination to identify the minimal set of
mutation(s) that are necessary and sufficient to confer antiviral resistance. Understanding which viral
protein or proteins mutate in order for the virus to escape inhibition suggests those proteins are targeted
by the therapeutic, and the site and phenotypic consequences of the mutations may provide insight into
the mechanism of antiviral activity. Together, this information provides a foundation for follow-up
structural, biochemical, and cell biological assays investigating the mechanism of antiviral activity. A
major strength of this approach is that it can be applied to any type of therapeutic, including therapeutics
with known targets (but unknown mechanisms of action) and therapeutics with unknown targets.
However, elucidating the mechanisms of antiviral activity based on indirect observations about antiviral
resistance can be challenging. For example, mutations may arise in proteins that are not directly targeted
by the therapeutic, or the phenotypic consequences of mutations may be unclear.732,733,734 Additionally, if
the drug targets a host protein, this approach provides indirect information about its mechanism of
activity.
Determining the Genetic Threshold for Resistance Development
Prior to the conduct of clinical trials and to support an IND application, the FDA recommends conducting
in vitro studies for selection of resistance to a therapeutic in order to determine the genetic threshold for
resistance development (i.e. how many mutations are needed to acquire resistance). Specifically, the FDA
recommends passaging the virus in the presence of therapeutic, followed by sequencing of emergent
resistant viruses and phenotypic characterization of resistant viruses.735 This experiment constitutes a GoF
approach.
9.10.3.5.3 Therapeutic Development Benefit #3: Inform Guidelines for Use of Therapeutics
The therapeutic regimen, including therapeutic dose and the use of combination therapies, may influence
whether and how readily antiviral resistance arises. Given that influenza viruses readily acquire resistance
to NAIs (i.e. upon acquisition of one or two mutations), influenza researchers cited a lack of knowledge
about the potential utility of combination therapies as a critical gap in public health preparedness for
influenza epidemics and pandemics.736 In addition, understanding whether antiviral resistance arises more
readily or differently in at-risk populations, such as obese or immunocompromised people, in either
scenario can provide information that further refines therapeutic guidelines. GoF approaches can address
each of these questions.
GoF approaches that lead to the development of viruses with resistance to therapeutics in development
can be used to evaluate the relationship between emergence of resistance and therapeutic dosage or the
732 Wensing AM et al (2014) 2014 Update of the drug resistance mutations in HIV-1. Topics in antiviral medicine 22: 642-650
733 Staschke KA et al (1995) Molecular basis for the resistance of influenza viruses to 4-guanidino-Neu5Ac2en. Virology 214:
642-646
734 Blick TJ et al (1998) The interaction of neuraminidase and hemagglutinin mutations in influenza virus in resistance to 4-
guanidino-Neu5Ac2en. Ibid. 246: 95-103

administration of multiple therapeutics in combination. First, serial passaging of virus in cells or animals
dosed with varying amounts of the therapeutic provides insight into the dose-dependence of the
emergence of resistant viruses. Second, serial passaging of virus in cells or in animals in the presence of
multiple therapeutics can be used to determine how readily resistance arises in response to combination
versus single therapies. Finally, serial passaging of virus in mouse models for at-risk populations (e.g.
immunocompromised mice or obese mice) provides additional information about the extent to which the
likelihood of resistance or patterns of resistance mutations vary depending on host factors, which may
inform therapeutic guidelines for specific at-risk populations.
9.10.3.6 Benefits and limitations of GoF Approaches to the Development of Diagnostics
GoF benefits to the development of therapeutics may have downstream economic benefits. Economic
9.10.4.1 Benefits and Limitations of alt-GoF approaches to Scientific Knowledge
Several GoF approaches enable the identification of mutations that confer antiviral resistance, including
serial passaging and targeted genetic modification to introduce mutations associated with antiviral
resistance. Conducting these approaches using attenuated reassortant strains in lieu of wild type strains
represents one type of alternative approach. Specifically, because experiments in this phenotypic category
focus on the influenza NA protein, reassortment strains containing the NA gene or the HA and NA genes
from a seasonal strain of interest and the remaining six or seven genes from the lab-adapted, attenuated
strain PR8 (7:1R or 6:2R strains) can be used. Influenza researchers felt that results about whether
mutations do or do not confer antiviral resistance in the context of attenuated reassortant strains are
generally reliable but cautioned that results may not be recapitulated in the context of the wild type
strain.738 Additionally, 6:2R and 7:1R strains cannot be used to discover or explore antiviral resistance
that arises due to mutations in virus proteins other than the NA (or HA) genes.
Several alternative experimental approaches can also be used to identify mutations that lead to antiviral
resistance. Comparative sequence analysis of wild type strains that are antiviral-resistant and antiviral-
sensitive enables identification of mutations that are associated with antiviral resistance. However,
because of the high genetic diversity among influenza viruses, identifying relevant mutations may be
difficult. One notable exception is comparative analysis of patient isolates over the course of antiviral
treatment, which is more readily able to identify mutations associated with antiviral resistance due to the

genetic similarity among patient isolates. However, the ability to opportunistically sample and analyze
patient isolates is likely to be relatively rare. Furthermore, both studies provide associative information.
Forward genetic screens to identify mutations that restore antiviral sensitivity to antiviral-resistant strains
(LoF) represents another alternative approach for discovering genetic traits linked to antiviral resistance.
Because this approach involves screening mutants, it is less efficient than GoF approaches for the
discovery of antiviral resistance traits, which rely on selection. Additionally, this approach is limited to
the study of antiviral-resistant strains that have arisen in nature and cannot be used to proactively identify
novel genetic traits that are associated with antiviral resistance. Targeted genetic modification of antiviral-
resistant strains to introduce mutations expected to restore antiviral sensitivity can be used to demonstrate
that a particular trait is necessary for antiviral resistance. Given that single mutations are typically
sufficient to confer resistance to NAIs, targeted LoF and GoF approaches are equally capable of
establishing a causal link between a particular genetic trait and antiviral-resistance. However, because use
of the targeted LoF method relies on the existence of an antiviral-resistant strain carrying a particular
resistance mutation of interest in nature, LoF is of limited utility for demonstrating that a resistance trait is
conserved across multiple strain contexts than its GoF counterpart.
The use of in vitro, virus-free systems represents another alternative approach for the study of genetic
traits underlying antiviral resistance. Several in vitro, virus free systems for the study of NAI resistance
have been used, which rely on ectopic expression of the influenza NA gene in cell culture.739,740 Using
these systems, forward genetic screens can be used to discover novel mutations that confer resistance.
Targeted mutagenesis of wild type NA genes can then be used to demonstrate that a particular mutation or
set of mutations is necessary and sufficient to confer resistance, as well as to determine whether the
phenotypic consequences of the mutation(s) are conserved across multiple genetic contexts. This
approach can be successfully used to study mutations that confer resistance by altering the function of the
NA protein. However, this approach cannot be used to uncover or to study mutations that confer
resistance by altering the expression levels of the NA protein, as has been documented for the H275Y
mutation (N1 numbering),741 or mutations in other genes that give rise to resistance through epistatic
effects. Additionally, given that antiviral-resistance is a continuum, results may not be recapitulated (or be
clinically relevant) in the context of the full virus.
Finally, computational models have been used to predict mutations that disrupt binding between NAIs
and the NA protein, which are expected to lead to antiviral resistance. While these models can be used to
generate hypotheses about antiviral resistance mutations in any virus strain, all predictions must be
experimentally confirmed through targeted mutagenesis, a GoF approach.
9.10.4.1.2 Scientific Knowledge Gap #2: What Selection Pressures Shape Whether and How Readily
Several alternative approaches can be used to gain insight into selection pressures that shape the evolution
and spread of antiviral resistance. Comparative analysis of the sequences and phenotypic characteristics
of patient isolates over the course of antiviral treatment has potential to provide direct insight into the
mechanisms driving emergence of antiviral resistance in people, including the identification of negatively
selected traits. However, as these studies are typically conducted in immunocompromised patients due to
their longer course of illness, results may not be representative of the general population. In addition,
739 Nivitchanyong T et al (2011) Enhanced expression of secretable influenza virus neuraminidase in suspension mammalian
cells by influenza virus nonstructural protein 1. Journal of virological methods 178: 44-51
740 Schmidt PM et al (2011) A Generic System for the Expression and Purification of Soluble and Stable Influenza
Neuraminidase. PLoS ONE 6: e16284
741 Bloom JD et al (2010) Permissive secondary mutations enable the evolution of influenza oseltamivir resistance. Science
(New York, NY) 328: 1272-1275

relative to animal passaging experiments (GoF), opportunities to conduct studies involving patients are
likely to be relatively rare due to ethical considerations.
Comparative analysis of the phenotypic properties (e.g. fitness) of antiviral-resistant and antiviral-
sensitive wild type strains can reveal genetic and phenotypic changes that are associated with the
acquisition of antiviral resistance, which may provide insight into the viral properties that shape the
evolution and spread of antiviral resistance in nature. However, the surveillance record is static and
cannot provide insight into negatively selected traits. Moreover, current surveillance efforts, which
largely involve consensus sequencing, are unlikely to capture the emergence of rare antiviral-resistant
variants. For these reasons, comparative analysis of wild type viruses provides limited insight into the
evolutionary mechanisms driving the evolution of antiviral resistance.
Other alternative approaches are not suitable for the study of evolutionary pressures that shape the
emergence and spread of antiviral resistance. In vitro, virus free approaches cannot provide insight into
how antiviral resistance affects other virus phenotypes, and current computational models cannot account
for epistatic effects (e.g. how antiviral resistance affects fitness). The use of attenuated reassortant strains
for GoF selection approaches, in lieu of wild type viruses, is of limited utility for studying the evolution
of antiviral resistance because the fitness of attenuated strains is altered relative to the wild type strains.
9.10.4.2 Benefits and limitations of alt-GoF approaches to surveillance
influenza viruses that have caused human infections. Resistance to NAIs can be assessed in two ways:
through laboratory testing of NAI resistance and/or by inspecting sequences for the presence of mutations
that are known to confer NAI resistance. As discussed in Sec. 9.10.3.2.2, the practice of using molecular
markers is constrained by several sources of scientific uncertainty, namely whether known markers are
conserved across multiple strain contexts and whether as-yet-unknown markers are capable of conferring
resistance. This section first reviews the ability of alternative experimental approaches to address those
scientific questions, then reviews the strengths and limitations of using phenotypic assays to assess the
antiviral sensitivity of surveillance isolates (i.e. relative to molecular markers).
9.10.4.2.1 Utility and limitations of alt-GoF approaches for strengthening the predictive value of
molecular markers for antiviral resistance
Alt-GoF approaches can be used to discover new mutations associated with antiviral resistance and to
validate known markers for antiviral resistance in different genetic contexts but have significant
limitations relative to GoF approaches. In vitro, virus-free systems can also be used to discover and
validate new mutations that give rise to antiviral resistance by altering the function of the NA protein, but
results may not be recapitulated in the context of the full virus. Targeted mutagenesis of antiviral-resistant
strains to restore antiviral sensitivity (LoF) can establish a causal link between a particular trait and
antiviral resistance, but the ability of this approach to demonstrate that particular markers are conserved
across multiple strain contexts is limited relative to GoF approaches because it relies on the existence of
antiviral resistant strains in nature. Comparing the sequences of wild type viruses or of patient isolates
over the course of antiviral treatment enables the identification of mutations that are associated with
antiviral resistance, which must be confirmed using targeted mutagenesis (GoF or LoF) to be useful for
surveillance. In addition, these approaches are limited to the discovery of antiviral resistance mutations
that have already arisen in nature. Computational models can be used to predict mutations that disrupt the
interaction between an NAI compound and an antiviral, but predictions must be validated experimentally.

9.10.4.2.2 Strengths and limitations of using phenotypic assays to characterize the antiviral sensitivity of
surveillance isolates
The strength of phenotypic assays, relative to predictive approaches, is that phenotypic assays provide a
direct readout of antiviral resistance. However, the practice of characterizing the antiviral sensitivity of
surveillance isolates through phenotypic assays has several shortcomings. These shortcomings were
discussed in detail in Sec. 9.6.3.2.1 and are briefly summarized here. First, the need for viral isolates
limits the number of viruses that can be subjected to phenotypic characterization. Second, the composition
of viral species present in the original clinical sample changes during isolation, as the most fit viral quasi-
species outcompete others. This change is of particular concern for antiviral resistance testing because
antiviral-resistant viruses are often less fit than antiviral-sensitive viruses, thus the presence of antiviral
resistant strains in mixed infections can be obscured as a result of virus isolation. Finally, delays in
shipping samples to WHOCCs for antiviral susceptibility testing, stemming from logistical, political,
and/or regulatory factors, create a lag time between sample collection and phenotypic characterization.
9.10.4.3 Benefits and Limitations of alt-GoF approaches to Decision-Making in Public Health Policy
9.10.4.3.1 Benefits to Decision-Making Related to Seasonal Flu Strains
recommendations to clinicians for the use of influenza antivirals. A subset of strains collected through
surveillance is subjected to phenotypic testing for antiviral resistance. Because phenotypic assays provide
a direct readout of the antiviral sensitivity of a given strain, phenotypic testing is likely to remain a critical
aspect of antiviral resistance monitoring in the future.
9.10.4.3.2 Benefits to Decision-Making Related to Pandemic Influenza
preparedness initiatives (discussed in detail in Sec. 9.6.3.3.3). As described above (Sec. 9.10.3.3.2), the
antiviral resistance element is important when coupled to other factors indicative of increased pandemic
potential, such as a high number of human infections or enhanced transmissibility or virulence in ferrets.
Importantly, when evaluating antiviral resistance, stakeholders consider both phenotypic and genetic data,
given the caveats associated with both types of data. Stakeholders emphasized that even following a rapid
risk assessment based on sequence data, confirming the antiviral resistance phenotype through phenotypic
testing is critical.
9.10.4.4 Benefits and Limitations of alt-GoF Approaches to the Development of Vaccines
9.10.4.4.1 Vaccine Development Benefit: Targeted Mutagenesis to Remove Antiviral Resistance Markers
from Vaccine Viruses
As discussed in Sec. 9.10.4.1.1, alt-GoF approaches are less efficient and effective than GoF approaches
for the discovery of novel viral virulence traits. However, targeted mutagenesis of antiviral-resistant
strains to introduce mutations expected to restore antiviral sensitivity (LoF), can be used to demonstrate
that a particular amino acid or set of amino acids are necessary for antiviral resistance. As the ultimate
goal is to restore antiviral sensitivity to vaccine strains harboring NAI-resistant NA genes, this approach
is as suitable as its GoF counterpart for identifying molecular markers linked to antiviral resistance for
this application.

9.10.4.5 Benefits and Limitations of alt-GoF Approaches to the Development of Therapeutics
GoF approaches are used to screen therapeutic candidates based on how readily antiviral resistance
emerges, and also to determine whether emergence of resistance enhances the infectivity, transmissibility,
or virulence of a virus, which is an important aspect of safety testing of the therapeutic candidate. No
alternative approaches can provide this information prior to field deployment of a therapeutic.

Candidates
Determining the Mechanism of Action of a Therapeutic
The FDA recommends submitting information about the mechanism of action of a therapeutic as part of
an IND application. This section reviews the ability of alt-GoF approaches to provide insight into the
mechanism of action of a candidate therapeutics.
Therapeutic candidates that are identified through high-throughput screens may attenuate viral replication
by directly targeting viral proteins or by indirectly targeting host proteins. For that reason, emergence of
resistance studies, which investigate potential viral targets, are usually complemented by high-throughput
RNAi screens targeting host proteins, to investigate potential host targets. Specifically, the fact that
knockdown of a particular host protein impedes the drug’s ability to inhibit viral replication suggests that
that protein or that signaling pathway may be targeted by the therapeutic. Though an informative strategy
for the study of therapeutics targeting host proteins, high-throughput RNAi screens provide minimal
information about potential viral targets of therapeutics.
a process in the viral life cycle may be difficult from such a static snapshot. Critically, because of the high
level of effort required for X-ray crystallography, it is not a feasible approach for simply screening the
potential viral targets of an unknown antiviral.
with a known target. In brief, this approach relies on the use of a “photoaffinity analogue” of the
candidate therapeutic, which is synthesized to contain a photosensitive group (e.g. an azide) and a
radioactive isotope (e.g. tritium, 3H).744 After treating the viral protein with the photoaffinity analog, the
sample is irradiated with UV light, triggering the photosensitive group to form a covalent bond with the
viral enzyme. Analytical techniques such as mass spectrometry can then be used to identify the labeled
amino acid residues in order to determine the drug’s binding site. This technique shares strengths and
weaknesses with X-ray crystallography. Namely, photoaffinity cross-linking is useful for small molecule
drugs that directly bind to and inhibit the activity of a viral protein and does not require prior knowledge
744 Cohen KA et al (1991) Characterization of the binding site for nevirapine (BI-RG-587), a nonnucleoside inhibitor of human
immunodeficiency virus type-1 reverse transcriptase. The Journal of biological chemistry 266: 14670-14674

of the location of the drug binding site.745 However, inferring the mechanism of antiviral activity based on
knowledge about the drug-virus protein interaction may be difficult.
Determining the Genetic Threshold for Resistance Development
resistance development (i.e. how many mutations are needed to acquire resistance). The FDA guidance
does not suggest any alternative approaches that could provide similar information. In fact, prior to
deployment of the therapeutic and the emergence of resistant viruses in nature, no alternative approaches
can provide this information.
GoF approaches can provide insight into the therapeutic dose that is least likely to lead to the emergence
of resistance as well as whether combination therapies are less likely to lead to the emergence of
resistance than single therapies. No alternative approaches are capable of providing similar information
about the dose-dependence of resistance or whether combination therapies lead to resistance less readily
than individual therapies.
approaches
GoF approaches are uniquely capable of identifying mutations that are necessary and sufficient to confer
antiviral resistance across multiple strain contexts, which provides a strong foundation for follow-up
studies to elucidate the mechanisms underlying antiviral resistance. GoF approaches also represent the
most efficient and effective approach for discovering novel mutations that confer antiviral resistance in
any virus strain, as conducting experiments with wild type viruses allows for discovery of the full
spectrum of mutations that may confer resistance, including mutations that alter the function or
expression level of the NA gene as well as mutations in other virus proteins that cause resistance through
epistatic effects. Attenuated reassortant strains may be used in lieu of wild type strains for many of these
experiments, but results may not be recapitulated in the context of the wild type viruses, particularly if
antiviral resistance arises through mechanisms other than changes to the function of the NA protein.
Alternative approaches can provide valuable insight into the study of antiviral resistance mechanisms but
have limitations relative to GoF approaches. Discovering new genetic traits associated with antiviral
resistance through comparative analysis of wild type sequences may be difficult. The comparison of
patient isolates over the course of antiviral treatment is a notable exception, but opportunities for such
studies are likely to be relatively rare. LoF approaches are relatively inefficient for the discovery of novel
genetic traits associated with antiviral resistance but can be used to demonstrate that a particular mutation
is necessary for antiviral resistance. Notably, the targeted LoF approach is often as capable of establishing
a causal link between a particular mutation and antiviral resistance as the targeted GoF approach because
NAI resistance is often conferred by single mutations; however, the ability of targeted LoF to demonstrate

that particular markers are conserved across strain contexts is limited by the number of of antiviral
resistant strains in nature. In vitro virus-free systems can be used to discover and validate mutations in the
NA gene that give rise to resistance but are not suitable for the study of resistance mechanisms that
involve alterations to gene expression levels or epistatic effects, and results may not be recapitulated in
the context of the full virus. Computational models may be used to predict novel mutations that confer
resistance by disrupting binding between the NAI molecule and the NA protein, but all predictions must
be experimentally confirmed using GoF approaches.
9.10.5.1.2 Scientific Knowledge Gap #2 – What Selection Pressures Shape Whether and How Readily
GoF approaches, namely serial passaging of viruses in animals in the presence of therapeutics, represent
the most efficient and effective strategy for gaining in-depth insight into the viral and host selection
pressures that shape the emergence and spread of antiviral resistance. Notably, attenuated reassortant
strains cannot be used for these studies because the phenotypic properties that are likely to shape the
likelihood that antiviral resistant strains will spread and persist in human populations, such as fitness, are
altered in these strains. While gaining direct insight into the behavior of the virus in humans through
human challenge studies (GoF) is valuable, these studies are rare due to ethical considerations.
Comparative analysis of patient isolates over the course of antiviral treatment can also provide in-depth
insight into the evolution of antiviral resistance in people, but studies are typically conducted in
immunocompromised patients and thus may not translate to healthy populations. Comparative analysis of
wild type isolates provides limited mechanistic insight into the viral or host factors that shape evolution of
antiviral resistance. Finally, neither virus-free approaches nor in silico approaches can be used to study
the interplay between antiviral resistance and other virus phenotypes.
influenza viruses that have caused human infections. Resistance to NAIs can be assessed in two ways:
through laboratory testing of NAI resistance and/or by inspecting sequences for the presence of mutations
that are known to confer NAI resistance. GoF approaches have the potential to benefit surveillance for
antiviral resistant strains by strengthening the predictive value of molecular markers for antiviral
resistance.
9.10.5.2.1 Benefits of GoF approaches relative to alt-GoF approaches for strengthening the predictive
value of molecular markers for antiviral resistance
GoF approaches represent the most efficient and effective methods for discovering novel mutations
associated with antiviral resistance and are uniquely capable of demonstrating that a particular mutation is
necessary and sufficient to confer antiviral resistance across multiple strain contexts. Alternative
approaches have significant limitations for the discovery of new markers and for the validation of known
markers. Taken together, GoF approaches provide unique benefits for strengthening the predictive value
of molecular markers for antiviral resistance, thereby improving their utility for interpreting surveillance
data.
9.10.5.2.2 Benefits of using molecular markers (GoF) versus phenotypic assays (alt-GoF) to characterize
the antiviral sensitivity of surveillance isolates
Both phenotypic assays and inspection of sequences for molecular markers of antiviral resistance have
strengths and limitations. Phenotypic assays provide direct information about the degree of antiviral

resistance of a particular strain, but results are delayed relative to sample collection and the properties of
viral isolates may not reflect the properties of viral quasispecies present in the original clinical sample.
For these reasons, researchers and government officials involved in influenza surveillance value the
ability to corroborate phenotypic assay data with sequence-based predictions based on molecular markers
of antiviral resistance, particularly when clinical samples can be directly sequenced. Of note, NAI
resistance markers that have been shown to be conserved across multiple strain contexts and are currently
incorporated into the practice of analyzing surveillance data.746 Thus, the benefits of GoF research about
molecular markers for antiviral resistance to the practice of surveillance can be realized immediately. As
sequencing becomes more common at NICs and other diagnostic laboratories and the number of known,
validated markers for NAI resistance rises, the molecular marker approach will take on relatively greater
importance. Ultimately, due to the rapidity of sequence-based analysis relative to phenotypic assays, the
use of molecular markers may increase capacity to monitor for antiviral resistance.
Notably, most genetic surveillance data is generated by sequencing of viral isolates at WHOCCs, though
the number of NICs and other diagnostic labs with sequencing capabilities is rising (though most of these
labs carry out sequencing on viral isolate samples). Full realization of the benefits that can be derived
from the use of molecular marker data will require an expansion of sequencing capabilities at diagnostic
laboratories that comprise the “base” of the influenza surveillance system as well as increasing the
number of clinical samples that are directly sequenced. Both capabilities have increased over the past
decade and are expected to continue to increase.747
9.10.5.3 Benefits to Decision-making in Public Health Policy
As described above, GoF approaches have potential to benefit surveillance for antiviral resistant strains,
which informs downstream decision-making related to public health practice and policy. Namely, data on
the prevalence of antiviral-resistant seasonal strains informs therapeutic recommendations developed by
the CDC, and antiviral resistance is one of the risk elements considered in a pandemic risk assessment of
an animal influenza virus.
Currently, a subset of the seasonal influenza viruses that are collected by WHOCCs are sent to CDC for
antiviral susceptibility testing.748 As discussed above, the use of molecular markers (GoF) has potential to
strengthen the robustness of antiviral resistance data, by corroborating phenotypic assay data, and to
increase the number of strains that can be phenotypically characterized. This expansion will provide a
stronger foundation for antiviral treatment recommendations and may enhance detection of rare antiviral-
resistant variants to increase awareness. However, given the inherent uncertainty of sequence-based
predictions, researchers and governmental officials involved in the analysis of surveillance data
emphasized that predictions should be validated through antiviral resistance assays whenever possible.
For pandemic risk assessments, the observation of antiviral resistance does not increase risk a priori but
rather is important when coupled to other factors indicative of increased pandemic potential, such as a
high number of human infections or enhanced transmissibility or virulence in ferrets. In this case, the
observation of antiviral resistance may trigger USG representatives to initiate the process of applying for
Emergency Use Authorization (EUA) from the FDA for antivirals in development, to ensure that effective
antivirals will be available if the strain under evaluation spreads to cause a pandemic. Importantly, when
evaluating antiviral resistance, stakeholders value the ability to corroborate phenotypic data with analysis
of genetic data, given the caveats associated with phenotypic assays. Additionally, the ability to conduct a
rapid risk assessment based on sequence data is valuable when strains first emerge and sequences are
746 (2015t) Interviews with influenza researchers and U.S. government representatives involved in influenza surveillance.

published prior to the receipt of viral isolates. Early decision-making can provide a several week head
start on the EUA process. As EUAs may be issued within days following a request if the FDA has worked
with government partners on a “pre-EUA” package, this head start could significantly impact the timing
of availability of antivirals in the event of a pandemic.749750,751
9.10.5.4 Benefits to the Development of Vaccines
Antiviral resistance markers can be mutated out of vaccine viruses to increase the safety of the vaccine
production process. Although GoF approaches represent the most efficient and effective way to discover
novel markers for antiviral resistance, either targeted GoF or LoF approaches can be used to establish a
causal link between a particular genetic marker and antiviral resistance, which is needed for translation of
this information to the vaccine production process. As the ultimate goal is to restore antiviral sensitivity to
vaccine strains harboring NAI-resistant NA genes, either GoF or alt-GoF approaches are equally suitable
for identifying molecular markers linked to antiviral resistance for this purpose.
9.10.5.5 Benefits to the Development of Therapeutics
GoF approaches are uniquely capable of screening potential therapeutic candidates based on how readily
antiviral resistance emerges as well as determining whether the emergence of resistance enhances the
infectivity, transmissibility, or virulence of a virus, an important aspect of safety testing of the therapeutic
candidate.

Candidates
The FDA recommends that an IND application include information about the mechanism of action of the
proposed therapeutic as well as selection for resistance studies to demonstrate the genetic threshold for
resistance to the therapeutic.
Serial passaging of a virus in the presence of therapeutic to discover mutations that confer resistance, a
GoF approach, is uniquely capable of identifying the viral target of a novel therapeutic with an unknown
mechanism of action. Given that researchers are undertaking unbiased screens to identify candidate
therapeutics that inhibit viral replication, this represents a valuable benefit for the development of new
influenza therapeutics. For therapeutics with known viral targets, this information about resistance
mutations can provide foundational information to guide follow-up structural, cell biological, and
biochemical studies investigating the mechanism of action of the therapeutic. Although crystallography
and photoaffinity cross-linking can also provide insight into the antiviral mechanisms of therapeutics that
directly bind to and inhibit virus proteins, inferring mechanistic information based on static information
about the virus-antiviral complex may be difficult. In these cases, knowledge about mutations that confer
resistance, generated through GoF approaches, provides an additional source of information that can be
used to generate testable hypotheses about mechanism of antiviral activity. Finally, the identification of
host factors that are required for antiviral activity is a critical aspect of examining therapeutics with
unknown targets. Though solely using host-focused approaches to elucidate the antiviral mechanism of a
749 Administration FaD. Guidance - Emergency Use Authorization of Medical Products.

http://www.fda.gov/RegulatoryInformation/Guidances/ucm125127.htm. Last Update Accessed November 10, 2015.
750 FDA. Emergency Use Authorization of Medical Products.
http://www.fda.gov/RegulatoryInformation/Guidances/ucm125127.htm. Last Update October 22, 2014. Accessed November
28, 2015.
751 (2015y) Personal communication from FDA representative.

therapeutic that targets the virus would be difficult, this information complements GoF approaches to
strengthen the evidence base for the drug’s mechanism of action.
GoF approaches are uniquely capable of determining the genetic threshold for resistance to a new
therapeutic, prior to field deployment of that therapeutic.
uniquely capable of determining the therapeutic dose that is least likely to lead to the acquisition of
antiviral resistance as well as determining whether combination therapies better prevent the emergence of
resistant viruses than individual therapies. Both types of information benefit the development of
therapeutic strategies that will be effective for a longer period of time in the field.
9.11 Influenza viruses: Benefits of GoF Research Involving Reassortment
9.11.1 Summary
This section describes the benefits of GoF studies that investigate the reassortment potential between two
viruses, which may lead to one or more of the phenotypic changes detailed in the NSABB Framework.
Such GoF studies were found to generate scientific knowledge, to inform surveillance of circulating
animal influenza viruses, which has downstream impacts on decision-making about USG investments in
pandemic preparedness initiatives such as pre-pandemic vaccine development, and to inform community-
level interventions aimed at preventing the emergence of novel reassortant viruses in human populations.
Alt-GoF approaches that may generate similar benefits were also identified and analyzed. At present,
GoF studies that involve reassortment have unique benefits to scientific knowledge and public health
practice, but realization of benefits to surveillance is subject to significant improvements to surveillance
capabilities for reassortant viruses. Chapter 9.11 provides an overview of these benefits, including basic
background and supporting information; a fully referenced and more thorough discussion of these
benefits can be found in Appendix IV Sec. 15.8.
9.11.1.1 Benefits of GoF studies that involve reassortment to scientific knowledge

o Uniquely capable of proactively assessing the potential for any two influenza viruses to
reassort and of comprehensively evaluating the viability of various gene combinations.
However, the outcomes of forced reassortment events in the laboratory may not reflect what
is possible or likely to occur in nature, and results from animal models may not translate to
human disease.
• Alt-GoF approaches are:

o Uniquely capable of studying co-infection and reassortment events in nature, but provide
limited mechanistic insight due to the complexities of the interactions between the viruses,
the host, and environmental factors that influence reassortment outcomes.
9.11.1.2 Benefits of GoF studies that involve reassortment to surveillance
• GoF approaches:
o May inform rapid assessment of the risks posed by circulating reassortant viruses detected
through surveillance, although the value of this benefit is limited by the fact that laboratory

results may not translate to viruses observed in nature. In addition, full realization of this
benefit will require significant improvements to surveillance capabilities for reassortant
viruses.
o May inform assessment of the risks posed by circulating reassortant viruses detected through
surveillance, although this information is subject to the availability of surveillance isolates
and will be delayed relative to the application of GoF data. In addition, full realization of this
benefit will require significant improvements to surveillance capabilities for reassortant
viruses.
9.11.1.3 Benefits of GoF studies that involve reassortment to public health practice and policy
• GoF approaches:
o GoF approaches may help to prioritize community-level interventions that aim to limit cross-
species interactions that would provide opportunities for co-infection between human
seasonal viruses and animal influenza viruses that have not yet infected humans. This benefit
arises from assessment of the potential and consequences of reassortment events, which is
one aspect of the risk posed by reassortment to human populations.
o GoF data may inform pandemic risk assessments of animal influenza viruses, which guide
downstream decision-making about pre-pandemic vaccine development and other pandemic
preparedness initiatives. GoF data are of low importance relative to other factors considered
in the risk assessment but can play an important role in rapid risk assessments when novel
viruses first emerge in human populations.
o Alt-GoF approaches may help to prioritize community-level interventions that aim to limit
cross-species interactions that would provide opportunities for co-infection between human
seasonal viruses and animal influenza viruses that have not yet infected humans. This benefit
arises from new insights into the ecological factors that shape the likelihood of reassortment
events occurring in nature, which is another aspect of the risk posed by reassortment to
human populations. Thus, this data complements that generated by GoF approaches to refine
prioritization of “prevention” activities.
o Alternative factors considered in the risk assessment, in particular epidemiological and

virologic factors, have a higher weight than the genomic variation risk element (informed by
GoF). However, these data may be scant or delayed relative to sequence data when novel
viruses first emerge in human populations.
9.11.2 Overview of GoF research that involves reassortment
This assessment describes the benefits of GoF experimental approaches that aim to assess the genetic
compatibility and fitness of viruses following reassortment. While the phenotypic consequences of
reassortment events between two viruses cannot be predicted with certainty, reassortant strains may
exhibit enhanced fitness, pathogenicity, and/or transmissibility relative to one or both parental strains.
(Notably, reassortant strains may also display reduced fitness, pathogenicity, and/or transmissibility
relative to parental viruses.) In this section, we provide an overview of GoF approaches that can be used
to assess the reassortment potential between two viruses and describe the scientific outcomes of each
approach.

9.11.2.1 Targeted reassortment by combining viral gene segments from two or more viruses to generate
viable reassortant viruses
Targeted reassortment of virus gene segments from two or more wild type virus isolates followed by
characterization of fitness in cell culture or in representative animal models is used to assess genetic
compatibility. This approach is in part performed to evaluate the genetic compatibility and viability of a
single reassortant virus, which can inform the understanding of the mechanisms underlying genetic
compatibility between virus gene segments across virus strains and subtypes. For example, a reassortant
virus compromised of virus gene segments sharing homology to the 1918 H1N1 pandemic virus from
eight different wild type avian isolates was generated to demonstrate that some 1918-like avian viruses
circulating in nature (which reassort frequently) are genetically compatible.752
9.11.2.2 Forward genetic screen to identify viable reassortant viruses
Forward genetic screens involve the generation of a panel of clonal recombinant viruses by
comprehensive reassortment of parental gene segments from two viruses (i.e. all or many possible gene
combinations), followed by characterization of the fitness of reassortants in appropriate mammalian
model systems. Follow-up studies may be performed to evaluate pathogenicity, infectivity, and/or
transmissibility of viable reassortants. This approach is performed to evaluate viability and genetic
compatibility of reassortant viruses, which provides a foundation for studies investigating mechanisms
governing reassortment and informs the potential for reassortant viruses to emerge in nature and the
potential public health consequences of such an emergence event.
9.11.2.3 Non-targeted reassortment using reverse genetics to select for viable reassortant viruses
In this approach, reassortants are generated using reverse genetics to mix viral gene segments of two wild
type viruses (i.e. mix up to 16 gene segments in total) in the context of cell culture or animal models. Use
of cell culture model systems involves the transient transfection of viral gene segments, while the in vivo
method involves the inoculation of ferrets with transiently transfected cells, followed by viral
reassortment in vivo. Both approaches are followed by limited passaging to select for viable reassortants.
Clonal isolates that emerge are then genotyped to identify their gene composition. This approach provides
insight into viable gene reassortment combinations as well as the relative fitness of reassortants under
selection pressures, which informs the potential and likelihood of reassortment emergence in nature.
9.11.2.4 Co-infection to select for viable reassortant viruses
In this approach, cultured cells or representative animal models are co-infected with two different wild
type viruses, followed by genotyping of clonal isolates that emerge during co-infection. This approach
determines the viability of various gene reassortment combinations and the relative fitness of reassortants
under selection pressures, which can inform the potential and likelihood of emergence in nature.
In this section, the potential benefits of GoF experiments that investigate the reassortment potential
between two viruses are discussed, in each benefit category listed in the NSABB Framework.
752 Watanabe T et al (2014a) Circulating avian influenza viruses closely related to the 1918 virus have pandemic potential. Cell
Host Microbe 15: 692-705

9.11.3.1 Scientific Knowledge
9.11.3.1.1 Background – critical gaps in scientific knowledge about reassortment
Here, the ability of GoF approaches to address a key outstanding question related to the reassortment of
influenza viruses in humans and other host species is evaluated:
• What is the potential for reassortment between two influenza virus strains?
o Are two influenza viruses genetically compatible?
o What is the relative fitness of reassortants that may affect the likelihood of their emergence
under selection in a host?
o How do selection pressures influence reassortment?
Reassortment involves the exchange of one or more complete virus gene segments between two different
viruses during the co-infection of a single cell, which can result in viruses that display enhanced fitness,
immune evasion and antigen escape, and resistance to antivirals.753 Considerable gaps in knowledge
remain about the biology of reassortment, including whether reassortment between two viruses will occur
and will lead to the generation of viruses with enhanced fitness, pathogenicity, and/or transmissibility.
GoF approaches can provide insight into these questions.
9.11.3.1.2 Benefits and limitations of GoF approaches that study reassortment
Several GoF approaches can lead to the generation of reassortant viruses:
• Targeted reassortment to generate a virus comprised of gene segments from two or more wild
type isolates,
• Forward genetic screens involving comprehensive reassortment to generate a panel of clonal viral
isolates followed by assessment of fitness in cell culture or representative animal models,
• Non-targeted reassortment involving gene segments from two different viruses to generate a
mixed population of reassortant viruses, followed by selection for compatible virus genotypes in
cell culture or representative animal models, and
• Co-infection of cell culture or representative animal models with two different viruses to select
for compatible virus genotypes.
Collectively, these approaches definitively demonstrate whether reassortment can occur between wild
type viruses and enable the identification of reassortment gene combinations that permit replication in in
vitro or in vivo model systems. This provides insight into the genetic compatibility of virus gene
segments. For the targeted reassortment approach, viral gene segments are selected based on a property of
interest (e.g. homology to a human pandemic virus) to answer a specific question about the genetic
compatibility between two or more viruses, which differs from the other GoF approaches that more
broadly query the range of reassortment combinations that are possible between two viruses.
753 Steel J, Lowen AC (2014) Influenza A virus reassortment. Current topics in microbiology and immunology 385: 377-401

Because forward genetic screens individually test every possible gene combination between two viruses,
this GoF approach can assess the viability and fitness of each viral clone that may otherwise be missed
with selection based approaches (below) in which more fit clones outcompete. However, the outcomes
associated with forward genetic screens are independent of the selection pressures that shape reassortment
potential and viral population diversity, and therefore may not fully represent the likelihood of
reassortants emerging.
The use of non-targeted reassortment by transfection of cell culture model systems with gene segments
from two separate viruses to select and identify emergent reassortants presents several different
advantages. First, this approach provides insight into how host pressures and competition among
reassortants shapes outcomes. Second, the ability to selectively remove a single gene segment that may
otherwise outcompete or skew virus populations enables assessment of the compatibility of many gene
segment combinations, relative to the co-infection approach.
Similar to the non-targeted reassortment approach, the co-infection approach provides insight into how
the host pressure and competition impact selection. A major benefit of this approach is that it mimics the
natural scenario under which reassortment can occur. However, in the event that two viruses of interest
display different tissue and cell tropism or significant disparity in fitness or infectivity, this approach
permits study of a limited number of reassortment combinations.
For all three approaches, the use of in vivo animal models for reassortment studies provides more relevant
information due to the complexity of host selection pressures relative to cell culture models. All GoF
approaches described here depend on whether the mechanisms and selection pressures underlying fitness
and reassortment in cell culture or animal models are representative of those in humans and whether the
genetic compatibility observed for the select number of strains analyzed is generalizable in other virus
contexts. Moreover, the use of the methods above may not capture the dynamics of co-infection and
reassortment in nature, limiting the relevance of results.
9.11.3.2.1 Surveillance for reassortant viruses – current process and limitations
The importance of reassortment in influenza virus biology is highlighted by its role in the emergence of
human pandemic viruses with minimal population immunity – all four of the influenza pandemics that
have occurred over the past century were likely caused by strains that arose through reassortment between
influenza viruses, although the role of reassortment in the emergence of the 1918 pandemic virus is
controversial.754,755,756,757,758 While the emergence of reassortant viruses cannot yet be predicted,
surveillance for reassortant viruses to assess their occurrence and prevalence in nature is of interest for
pandemic preparedness, and as such is one of the factors considered in pandemic risk assessments
(discussed further below). Determining whether a reassortant virus poses an increased risk to human
populations relative to its parental viruses poses a major challenge.
754 Morens DM, Fauci AS (2007) The 1918 influenza pandemic: insights for the 21st century. The Journal of infectious
diseases 195: 1018-1028
755 Belshe RB (2005) The origins of pandemic influenza--lessons from the 1918 virus. The New England journal of medicine
353: 2209-2211
756 Worobey M et al (2014) Genesis and pathogenesis of the 1918 pandemic H1N1 influenza A virus. Proc Natl Acad Sci U S A
111: 8107-8112
758 Smith GJ et al (2009b) Dating the emergence of pandemic influenza viruses. Proc Natl Acad Sci U S A 106: 11709-11712

Analysis of the phenotypic properties of reassortant viruses in a laboratory setting, in particular fitness,
pathogenicity, and transmissibility, provides insight into the properties associated with viable reassortants
and can call attention to particular reassortant viruses that display phenotypic properties of concern. This
information can inform evaluation of the risk posed by particular reassortant viruses detected in nature.
9.11.3.2.2 Benefits and limitations of GoF approaches to surveillance for reassortant viruses
GoF approaches that proactively determine the reassortment potential between two viruses and
phenotypic properties of reassortant viruses represent an efficient method for generating a breadth of
information that can inform rapid analysis of surveillance data. However, whether laboratory results
translate to the field strains of interest in nature is uncertain, given differences in the genetic sequences of
the laboratory and field strains and the inherent artificiality of studies conducted in model systems in a
laboratory setting.
9.11.3.3 Decision-making in public health practice and policy
GoF reassortment studies have potential to benefit two aspects of public health practice and policy. First,
the results of reassortment studies may stimulate risk mitigation activities to limit the potential for risky
co-infections to occur in nature in human and/or animal hosts (i.e. those co-infections that could give rise
to reassortant viruses with risky properties). Second, reassortment studies may inform pandemic risk
assessments of circulating animal influenza viruses, which guide downstream decision-making about pre-
pandemic vaccine development and other pandemic preparedness initiatives.
9.11.3.3.1 GoF benefits to risk mitigation activities that aim to prevent the emergence of reassortant
viruses in nature
Reassortant viruses arise in nature during co-infection of a host with two different viruses. Limiting the
interaction between two different species can mitigate the risk of co-infection of either host with an
adapted and an “exotic” strain (e.g. co-infection of a human with seasonal H1N1 and avian H5N1), which
could give rise to a reassortant strain comprised of viral gene segments from both strains.759 GoF
approaches that proactively study the reassortment potential between two virus strains adapted for growth
in different species provides insight into reassortants that are viable and that display phenotypic properties
of concern. This information can help to prioritize risk communication about measures to mitigate the
chance of co-infections.760 For example, hunters would be encouraged to wear personal protective
equipment while gutting birds in areas where avian viruses that are capable of reassorting with human
seasonal viruses are circulating in game bird populations.761 Furthermore, data from GoF reassortment
studies provides an evidence base for messaging that may increase “buy-in” among the target population.
The results of GoF reassortment studies may also inform biosecurity practices at farms, with respect to
interactions between farm workers and animals, interactions between different species of animals (e.g.
poultry and swine at mixed-species farms), and interactions between agricultural animals and wildlife.
Environmental conditions that provide opportunities for co-infections with a human seasonal virus and an
animal flu virus that has already caused human infections are of high concern regardless of results from
laboratory reassortment studies.762 Thus, GoF studies that investigate the reassortment potential between
759 (2015q) Interviews with researchers at the National Wildlife Health Center (United States Geological Survey, Department of
the Interior).
the Interior).
762 Zhu Y et al (2013a) Human co-infection with novel avian influenza A H7N9 and influenza A H3N2 viruses in Jiangsu
province, China. Lancet 381: 2134

human seasonal viruses and animal viruses that have not yet caused human infections are likely to have a
larger impact on public health practice (e.g. HPAI H5N2 and human seasonal viruses).
9.11.3.3.2 GoF benefits to pandemic risk assessments and downstream decision-making for pandemic
preparedness
Pandemic risk assessments of circulating animal influenza viruses inform decision-making about how to
invest in public health preparedness activities for influenza pandemics, particularly development of pre-
pandemic vaccines. The genomic variation risk element of the Influenza Risk Assessment Tool (IRAT)
used by the USG for pandemic risk assessments, described in detail in Sec. 9.6.3.3.2, includes
consideration of reassortment. Specifically, reassortment between different lineages or sub-types of
viruses raises the risk score for this element. GoF approaches that provide insight into the properties of
reassortant viruses, in particular their fitness, transmissibility, and virulence, could be used to refine the
scores associated with this risk element. In this way, GoF approaches may benefit downstream decision-
making in public health policy.
In general, the genomic variation risk element is of low to intermediate importance relative to other
factors considered in the risk assessment, such as the number of human infections and the phenotypic
properties of the virus. Notably, corroboration of phenotypic data adds value to the assessment by
increasing certainty in downstream decisions. Furthermore, as discussed in detail in Sec. 9.6.3.3.3, the
genomic variation risk element may play a relatively more important role in the assessment when a novel
virus first emerges in human populations, if sequences are published prior to the shipping of viral isolates
to the U.S. The ability to evaluate risk based on genetic sequence data enables a rapid risk assessment,
which may trigger the decision to develop a CVV, providing a head start on vaccine production that
would be valuable in the event of a pandemic.
9.11.3.4 Vaccines, therapeutics, and diagnostics
GoF-derived information about the reassortment potential of two different viruses is not relevant for the
development of vaccines or therapeutics.
As existing influenza diagnostics are not equipped to rapidly screen and detect reassortants, information
about reassortants with phenotypic properties of concern could, in principle, guide development of
diagnostics to detect those reassortants. However, GoF approaches do not provide insight into the
likelihood that reassortment will occur in nature, which is a function of complex ecological factors that
govern the likelihood of co-infections. The likelihood of reassortment is also a critical factor for the
design of targeted diagnostics for reassortant viruses (i.e. there is no need to design diagnostics for rare
reassortant events). For this reason, GoF approaches are unlikely to trigger the development of new
diagnostics independently of the observation of co-infection or reassortment events occurring in nature.
No economic benefits of GoF reassortment studies were identified
9.11.4.1 Scientific knowledge
A select number of alt-GoF approaches can be used to analyze the reassortment potential of two different
viruses. Analyzing the sequences of human and animal surveillance isolates to detect reassortment events

can provide insight into the occurrence and prevalence of reassortment in nature. This approach includes
sequence inspection for several different types of reassortment events, involving:
• Two different human seasonal virus sub-types (e.g. H1N1 and H3N2),
• Human or animal virus strains within the same sub-type (e.g. different clades of H3N2),
• Human and animal viruses (e.g. human seasonal H3N2 and swine-origin H1N1), and
• Two different animal virus sub-types (e.g. H9N2 and H7Nx).
Analysis of both animal and human isolates provides information that is applicable to a broad number of
strains, and the analysis of human isolates provides information about reassortment potential that is
directly relevant to human populations. However, this approach is significantly limited by the quality and
availability of existing genetic surveillance data. In addition, this analysis is limited to the study of
reassortant viruses that have evolved (and have been subsequently detected) in nature.
A second type of alt-GoF approach involves the analysis of viral isolates from humans or animals that
have been co-infected with two influenza viruses. This approach can determine whether reassortment has
occurred and also may provide insight into the genetic compatibility of various gene combinations, as
well as host selection pressures that shape the outcome of reassortment events. That analysis of human
and animal isolates provides information that is directly relevant to reassortment potential in nature is a
strength of this method. However, this approach is also subject to significant limitations. Although co-
infection events occur, these events are captured on an ad hoc basis, thus opportunities for such studies
are likely to be relatively rare. Moreover, unknowns in the route of infection, the level and time of
exposure, and diversity in the host response due to existing natural or induced immunity limits the ability
of this approach to reliably assess genetic compatibility of reassortant viruses.
The use of replication incompetent viruses provides another alternative method for the analysis of genetic
compatibility between gene segments from two influenza viruses. In these model systems, viral
replication can be assessed in cell culture lines that are engineered to stably express an essential viral
protein that is missing from the “replication-incompetent” virus strains used for infection. The result is a
virus that is biologically constrained to replication in that cell line. Several replication incompetent model
systems have been made, and these systems have been used to assess the of genetic compatibility of
specific virus gene segments by targeted.763,764,765 However, this system has not yet been used to broadly
assess the reassortment potential between two viruses (i.e. reassortant viruses that emerge following
transfection of cells with all eight gene segments from both viruses, which mimics a co-infection event).
One major drawback is that this approach does not capture the complex selection pressures observed in
vivo. Additionally, results may not translate to reassortment in humans, and findings may not be
generalizable to other virus contexts.
A final alt-GoF approach utilizes in vitro virus-free methods to investigate genetic compatibility of viral
gene segments in isolation. In particular, forward genetic screens can be used to identify novel gene
segment combinations or reassortment events that contribute to a phenotype underlying viral fitness and
infectivity, such as polymerase activity. Though the simplicity and relatively high-throughput nature of
these methods renders them appealing as a screening approach for the evaluation of genetic compatibility
between two viruses, these approaches are inherently limited to the characterization of phenotypes
the Detection of Influenza Virus Neutralizing Antibodies. Journal of virology 84: 2157-2163
765 Baker SF et al (2014) Influenza A and B virus intertypic reassortment through compatible viral packaging signals. Journal
of virology 88: 10778-10791

previously identified in other experiments. In addition, results may not be recapitulated in the context of
the full virus or in vivo.
Characterization of field viruses, an alt-GoF approach, provides direct insight into the phenotypic
properties of reassortant viruses of interest. However, this approach is reactive and depends on the
availability of viral isolates or the publication of a high-quality, complete genome sequence for synthetic
reconstruction of the virus. Additionally, this approach provides limited mechanistic insight into the
relative fitness of reassortant and parental viruses, due to the high genetic diversity among circulating
influenza viruses. Finally, whether results gleaned from studies in laboratory animals translate to human
disease is uncertain.
9.11.4.3.1 Alt-GoF benefits to risk mitigation activities that aim to prevent the emergence of reassortant
viruses in nature
could give rise to a reassortant strain.766 Understanding whether reassortment between two viruses has
potential to generate viruses with phenotypic properties of concern (e.g. enhanced transmissibility,
virulence, etc.) can inform prioritization of community-level interventions that aim to limit opportunities
for “risky” co-infection events. Because alternative experimental approaches are reactive, limited to study
reassortment events that have already occurred in nature, these approaches have limited ability to inform
such proactive “prevention” initiatives.
However, the risk posed to human populations by reassortment events also depends on the likelihood that
co-infections and subsequent reassortment occurs. The likelihood of reassortment in nature depends on
complex ecological factors such as the distribution of viruses within and among reservoir species, which
are poorly understood. These factors can be studied using alternative approaches such as characterizing
the prevalence and distribution of influenza viruses circulating within and between animal reservoir
species. This information can provide insight into the factors that drive reassortment events in nature,
which will help to refine risk communication and community-level intervention efforts that aim to prevent
the emergence of novel influenza viruses in human populations through reassortment.
9.11.4.3.2 Alt-GoF benefits to pandemic risk assessments and downstream decision-making for pandemic
preparedness
Pandemic risk assessments of circulating animal influenza viruses inform decision-making about whether
and how to invest in pre-pandemic vaccine development and other pandemic preparedness initiatives.
the Interior).

The genomic variation element of the IRAT includes consideration of reassortment, which may be
informed by GoF studies that proactively assess the phenotypic consequences of reassortment events. In
addition to genomic variation, several other types of information related to the properties of the virus are
considered in the risk assessment: phenotypic data (i.e. transmissibility and virulence in ferrets),
epidemiological data (i.e. the number and severity of human infections), and ecological data (i.e. factors
related to infections in animals). In general, these factors are more important than the genomic variation
risk element, in particular epidemiological and virologic data. However, a major drawback associated
with these two data sources is that when novel viruses first emerge in human populations,
epidemiological data may be scant and virus shipping delays will delay the generation of virologic data.
approaches
9.11.5.1 Scientific knowledge
GoF approaches are uniquely capable of proactively assessing the potential for any two influenza viruses
to reassort, as well as for comprehensively evaluating the viability of various gene combinations. Notably,
the outcomes of forced laboratory reassortment events may provide limited insight into the likelihood that
such reassortment events will occur in nature, as natural reassortment depends on complex factors such as
the rate of co-infection and the distribution of genetically compatible viruses (which are unknown). In
addition, the relevance of this information for human populations depends on the suitability of animal
models. Although surveillance-based approaches can provide broad insight into of the prevalence and
distribution of reassortment viruses in different host populations, their utility is severely limited by the
quality and availability of surveillance data. Similarly, the analysis of humans or animal isolates during
co-infection is an unreliable method for determining the reassortment potential and genetic compatibility
of two viruses, and opportunities for such studies are rare. The use of replication incompetent viruses is a
promising approach for assessment of the genetic compatibility and reassortment potential between two
viruses, but this system is not commonly used for this purpose and requires further validation. Moreover,
it cannot capture the complex selection pressures observed in vivo and may not translate to mechanisms of
reassortment in humans. Although the use of in vitro virus free systems is useful from an initial screening
approach, results may not be recapitulated during the complete viral life cycle.
Both GoF and alt-GoF approaches provide information about the phenotypic properties of reassortant
viruses detected through surveillance, which can inform analysis of their potential risks to human
populations. The proactive nature of GoF studies facilitates more rapid assessment of surveillance data,
but results may not translate to the strains observed in nature. In contrast, alt-GoF approaches provide
more relevant information by directly studying the surveillance strains of interest but generate
information after strains have been detected and require a viral isolate or high-quality genetic data for
synthetic reconstruction of the virus.
Notably, the benefit of using experimental data about reassortant viruses (both GoF and alt-GoF) to aid
the interpretation of surveillance data is severely constrained by the quality and availability of existing
genetic surveillance data. Reassortment events are most commonly identified through individual
phylogenetic analysis of each viral gene segment to identify its origin and ancestry. 767 This requires full
genome sequences and large sequence databases for effective determination of phylogenetic ancestry,

which are not always available, particularly for influenza viruses isolated from animal reservoirs.768 Given
these limitations, GoF and alt-GoF approaches to study reassortment currently provide minimal benefits
to the interpretation of surveillance data. Full realization of their potential benefits will require significant
expansion of genetic surveillance for reassortant viruses.
9.11.5.3.1 Benefits to risk mitigation activities that aim to prevent the emergence of reassortant viruses in
nature
GoF studies that proactively study the reassortment potential between human seasonal viruses and animal
viruses that have not yet caused human infections may help to prioritize risk communication and risk
mitigation measures that aim to limit cross-species interactions that would provide opportunities for co-
infection. These data also provide an evidence base for risk mitigation messaging that may increase buy-
in among the target population. Alternative approaches can provide insight into the ecological factors that
drive reassortment in nature, which is also needed to refine prioritization of risk communication and
mitigation activities.
As environmental conditions that provide opportunities for co-infections with a human seasonal virus and
an animal virus that has caused human infections are already of high concern, reassortment studies
involving these viruses are unlikely to further increase preventive measures that are already in place.
9.11.5.3.2 Benefits to pandemic risk assessments and downstream decision-making for pandemic
preparedness
pandemic vaccines. The genomic variation risk element considered during pandemic risk assessments
may be informed by GoF studies involving reassortment. In general, the genomic variation risk element is
of low to intermediate importance relative to other factors considered in the risk assessment, in particular
epidemiologic and virologic factors. However, corroboration of phenotypic data adds value to the
assessment by increasing certainty in downstream decisions. Furthermore, GoF data plays a relatively
more important role when novel viruses first emerge in human populations, when epidemiological data
are likely to be scant and virus shipping delays will delay the generation of virologic data. In this case, the
ability to evaluate risk based on genetic sequence data enables a rapid risk assessment, which can provide
a head start on downstream response activities that would be valuable in the event of a pandemic.
9.12 Evaluation of the quantitative benefits of GoF research
This section quantitatively explores the benefit of GoF and alt-GoF experiments that influence the
availability of influenza vaccines during seasonal influenza epidemics and influenza pandemics. These
benefits are briefly summarized below and are described in detail in the relevant GoF phenotype section.
768 Vincent A et al (2014) Review of influenza A virus in swine worldwide: a call for increased surveillance and research.
Zoonoses and public health 61: 4-17

9.12.1 Overview of GoF and alt-GoF benefits subject to quantitative analysis
9.12.1.1 GoF experiments that enhance virus production
GoF approaches that enhance virus production are currently used to produce egg- and cell-based influenza
vaccines, which comprise over 99% of influenza vaccines produced annually in the U.S. Eliminating GoF
approaches from the current vaccine production process would likely result in the inability to produce
vaccine or the production of completely ineffective vaccines (due to poor vaccine match), as no
alternative approaches can supplant the use of GoF approaches in the near-term.
GoF approaches that enhance virus production can also improve the current influenza vaccine production
process. Specifically, GoF-derived improvements to the yields of vaccine viruses will increase the rate of
bulk antigen production, thereby shortening vaccine production timelines. The production of influenza
vaccines is highly optimized, such that current production capacities of eggs, the medium used for the
majority of flu vaccine production, are at or near maximum levels. As a result, benefits derived from
increasing vaccine virus yields primarily benefit vaccines based on viruses that grow poorly in eggs, such
as the 2009 H1N1 pandemic virus. That is, incorporating insights from GoF research into those initially
low-yield vaccine viruses could boost their production to “normal” levels. This improvement will lead to
faster vaccine availability during future pandemics caused by viruses that have naturally low yields in
eggs.
9.12.1.2 GoF experiments that enhance infectivity, transmissibility, and virulence in representative
animal models
GoF approaches that enhance the infectivity, transmissibility, and virulence of influenza viruses in
representative animal models also have potential to improve vaccine availability during a pandemic.
Specifically, these GoF approaches strengthen the predictive value of molecular markers for mammalian
adaptation, transmissibility, and virulence, which inform pandemic risk assessments of circulating animal
influenza viruses. These assessments guide downstream decision-making about investments in pre-
pandemic vaccine development, namely decisions about whether to develop candidate vaccine viruses
(CVVs), develop a vaccine seed strain, produce clinical lot material, conduct clinical trials, and stockpile
bulk antigen. In the event that a similar virus emerges to cause a pandemic, each of these preparative steps
will shorten the time needed for large-scale production of that vaccine. Developing pre-pandemic CVVs,
vaccine seed strains, and conducting clinical trials to determine the dosage, need for adjuvants, and other
dosing parameters will eliminate steps from the production process, and manufacturers’ experience
working with the vaccine strain will streamline the subsequent production process. These improvements
will translate to faster vaccine availability during a pandemic.
9.12.1.3 Alternative approaches that influence the availability of vaccines
Alternative approaches also have potential to increase the availability of influenza vaccines during a
pandemic. Several alternative approaches can shorten production timelines for strain-specific influenza
vaccines. First, the development of modified host cell lines that permit higher levels of virus replication
increases the rate of bulk antigen production. Second, incorporating adjuvants into existing egg- and cell-
based vaccines enables a smaller quantity of antigen to be used in each vaccine dose, thereby shortening
the overall production timeline. Third, new, virus-free vaccine platforms, such as recombinant vaccines,
have shorter production timelines than egg- and cell-based vaccines. Due to regulatory barriers, none of
these alternatives have potential to influence vaccine production timelines in the near term, but each has
potential to shorten production timelines in the intermediate- to long-term.

9.12.2 Overview of GoF benefits not subject to quantitative analysis
9.12.2.1 Influenza viruses
Other benefits of GoF research involving influenza viruses are not amenable to a meaningful quantitative
analysis.
Approaches within two phenotypic categories (enhanced virus production and evasion of existing natural
or induced adaptive immunity) have potential to improve the efficacy of seasonal flu vaccines. GoF
approaches that enhance virus production can shorten vaccine production timelines, enabling selection of
strains closer to the start of flu season, which will increase the likelihood that the “correct” strains are
chosen resulting in well-matched vaccines. GoF approaches that lead to evasion of existing natural or
induced adaptive immunity have potential to improve strain selection capabilities through several
different mechanisms, which will similarly increase the likelihood that vaccines are well-matched to
circulating strains at their time of deployment. The degree to which either advance will improve the
likelihood of vaccine match is highly uncertain. Furthermore, the relationship between vaccine match and
vaccine efficacy for a given flu season is complex, arising not only from the antigenic relationship
between the vaccine strain and the dominant circulating strain but also historical factors such as the
antigenic relationship between the dominant and recently circulating strains, vaccine coverage during the
current and past flu seasons, and other factors. Thus, determining how an assumed increase in vaccine
match translates to an increase in vaccine efficacy is also subject to considerable uncertainty. Given these
uncertainties, quantitatively assessing the benefits of GoF improvements to vaccine efficacy in a
meaningful way is not possible.
Approaches within several phenotypic categories (evasion of vaccines in development, enhanced

virulence, and evasion of therapeutics) may benefit the development of novel vaccines and therapeutics.
Exactly which novel medical countermeasures these studies may lead to is unknowable, so, even though
the benefit of novel countermeasures could be assessed parametrically, the advent of any countermeasure
with a specific property could not be tied directly to one of these GoF studies.
Finally, GoF studies involving reassortment may stimulate implementation of activities that aim to
prevent the emergence of novel flu viruses in human populations. Whether these activities prevented the
emergence of a pandemic virus is unknowable, thus this benefit was not quantified either.
The benefit associated with novel countermeasures against the SARS and MERS coronaviruses was not
assessed because two counterfactuals must be assumed. Firstly, no pandemic of these diseases has
occurred due to their susceptibility to public health control measures and spontaneous social distancing,
and so the effects of notional countermeasures on mitigating a pandemic have doubtful relevance.
Secondly, the geographic and temporal origins of the next outbreak of a novel strain of coronavirus are
unpredictable, and therefore, even if the medical countermeasures existed, the supply of these
countermeasures and the ability of the local public health system to distribute them is unknowable. For
this reason, the magnitude of the benefit of medical countermeasures to controlling a local outbreak
caused by a novel coronavirus is also subject to irreducible uncertainty.
9.12.3 Benefit Associated With Seasonal Influenza Vaccine
As described above, eliminating GoF from current vaccine production processes is likely to result in
production of a completely ineffective influenza vaccine due to poor vaccine match or the inability to
produce influenza vaccine. Figure 9.4 shows the number of deaths suffered in a typical seasonal influenza

outbreak given normal production and administration of vaccine compared to the complete absence of an
effective vaccine. Although administration of seasonal influenza vaccine doses begins just prior to the
start of “influenza season”, many influenza infections exist in the US before this time and the overall
predictions of deaths suffered is sensitive to how many infections are presumed to exist at this point. The
figures below presume that either 100, 1,000 or 10,000 people infected with seasonal influenza exist prior
to the onset of the season. Here, parameter values were chosen to illustrate models whose results match
those seen for average seasonal flu outbreaks in the USA, to more closely illustrate the predicted benefits
of seasonal flu vaccines. No matter which assumption is made, the lack of a vaccine significantly
exacerbates the outbreak, increasing the number of deaths by ten to 100 fold. The effect observed is due
not only to the protection of vaccinated individual from infection, but also greatly reduced case numbers
overall due to herd immunity dampening the outbreak. This finding demonstrates that any measures that
imperil vaccine production could have a significant and real cost.
Figure 9.4. The cost of losing an effective seasonal influenza vaccine. This figure shows the number of deaths
suffered in North America from a typical seasonal influenza outbreak given normal production and
administration of the influenza vaccine or in the absence of the vaccine. The three panels show the number of
deaths predicted if 100, 1,000 or 10,000 cases of influenza exist in North America at the time influenza
vaccines begin to be administered just prior to the start of “influenza season”. The R 0, case fatality rate,
community mitigation strength, vaccine efficacy, and antiviral distribution values were fixed at single values
illustrating an average flu season in the USA, and vaccine distribution was started immediately after the
simulations began. Lines represent the middle 80% of the results across all remaining varied parameters.
9.12.4 Benefit Associated with Pandemic Influenza Vaccine
Unlike seasonal influenza, outbreaks of pandemic influenza are currently unpredictable. Because of this
lack of warning, the length of the production cycle of vaccine to mitigate the pandemic is critical. The
production timeline of a pandemic influenza vaccine influences how quickly after the outbreak is detected
that the vaccine will be available. Figure 9.5 explores the relationship of the timing of the availability of
pandemic influenza vaccine and deaths in North America. The particular strain modeled has a death rate
and transmissibility that exceeds that of the 2009 pandemic strain (to better reflect other pandemic
strains). These data can be used to evaluate the quantitative benefits of several GoF and alt-GoF
approaches that influence the production timelines of influenza vaccines: (1) GoF approaches that are
currently used for vaccine production, which are needed to maintain the current ability to produce
pandemic influenza vaccines, (2) GoF approaches that shorten existing vaccine production timelines,
which have potential to improve vaccine availability in the near-term, and (3) alt-GoF approaches that

shorten vaccine production timelines (i.e. development of modified cell lines, use of adjuvants for dose
sparing, and development of new, virus-free vaccine platforms), which have potential to improve vaccine
availability in the intermediate- to long-term.
As described above, eliminating GoF approaches from existing vaccine production practices would likely
result in production of a completely ineffective vaccine or in the inability to produce a vaccine. The
consequences of having no vaccines available during a pandemic, relative to vaccines that can be
deployed on current production timescales, are illustrated through comparison of the “current typical” and
“never” timepoints on the graphs in Figure 9.5. As demonstrated, the benefit at mitigating the outbreak
for the vaccine depends heavily on how the public reacts to the outbreak. On the right, if the public barely
changes its behavior during the outbreak, current vaccine production timelines are too slow to prevent a
significant number of deaths, thus the difference between deploying vaccines on current timescales and
never deploying vaccines is minimal. However, if the public reacts strongly and reduces their usual
contacts by 25% (left panel, community mitigation 0.25), then any delay of vaccine production could
increase the number of deaths expected. The inability to deploy vaccine would increase the number of
deaths, relative to deployment of vaccine on typical timescales, 1.2 to six-fold depending on how rapidly
the rapidly the pandemic is detected (indicated by the number of seed infections at the start of vaccine
production).
GoF and alt-GoF approaches also have potential to shorten vaccine production timelines, which would
enable deployment of vaccine earlier during a pandemic. However, the extent to which alternative
approaches could shorten vaccine production timelines in the future is uncertain. If the public does not
change their behavior during a pandemic, production timelines must be shortened by more than six weeks
to significantly reduce the number of deaths (i.e. the production timeline must be faster than the ‘current
optimal’ timeline). If the public reduces their usual contacts by 25%, then any reduction in the time
needed to produce vaccines would reduce the number of deaths during a pandemic.
As described above, GoF approaches that enhance virus production will primarily aid production of
vaccines based on “slow” growing viruses, allowing these vaccines to be produced on closer-to-typical
timescales. Thus, comparison of the “current slow” and “current optimal” timepoints on the graphs in
Figure 9.5 provides an estimate of the scale of this benefit using a vaccine with mean efficacy during a
pandemic with median R0 and case fatality rate twice that of the seasonal outbreak above. If the public
does not change their behavior during the pandemic (right graph), this improvement to production would
have minimal impacts on the number of deaths because the typical production timeline is too slow for
vaccination to significantly mitigate the consequences of a pandemic. However, if the public reduces
contacts by 25% (left graph), the number of deaths predicted will decrease by roughly 30%.
Implementing one or more stages of the pre-pandemic vaccine development pipeline, influenced by GoF
approaches that enhance the infectivity, transmissibility, and virulence of influenza viruses, could also
shorten vaccine production timelines during a pandemic. Even if the public does not change their
behavior during the pandemic, shortening production timelines by nine weeks could reduce the number of
deaths by 15 to 30% (compare “current optimal” to “9 weeks faster” timepoints, right graph). If the public
does reduce contact rates, this improvement to production would decrease the number of deaths by 60 to
70%, which would save more than 100,000 lives in a high mortality outbreak.

Figure 9.5. The relationship between the timing of the availability of a vaccine against an emergent pandemic
influenza strain and deaths suffered in North America for two different values of community mitigation
strength. Results are shown for a vaccine of mean efficacy and an outbreak with a median R0and case fatality
rate twice that of the seasonal outbreak depicted above. The right panel shows results if the public barely
chances its behavior (10% fewer contacts) whereas the left panel shows the results if the public reduces its
contacts by 25% for the duration of the outbreak. The three lines on each graph show the results if the
vaccine production process starts when there are 100, 1,000 or 10,000 cases in North America.
9.13 Likelihood of GoF Strains Arising in Nature
9.13.1 Summary
GoF experiments that enhance the transmissibility or virulence of influenza viruses, that lead to evasion
of existing natural or induced adaptive immunity, and that lead to evasion of therapeutics are pursued to
gain insight into the mechanisms underlying those phenotypic changes and to generate information that
can benefit public health. Both the potential benefits of those experiments, as well as the public health
risks of not conducting the experiments, depend on the likelihood that the phenotypic changes observed in
the laboratory will occur in nature. Antigenic drift of seasonal influenza viruses and evolution of antiviral
resistance (in both seasonal and animal influenza viruses) both occur regularly in nature. Influenza viruses
exhibit a wide spectrum of virulence in humans. Notably, the 1918 H1N1 pandemic virus caused a case
fatality rate several orders of magnitude higher than seasonal influenza viruses, demonstrating that viruses
with high virulence can emerge to cause pandemics.
Animal influenza strains are not known to have directly evolved the capacity for efficient transmission in
humans. In contrast, the fact that the four influenza pandemics of the past century were caused by
reassortant viruses definitively demonstrates that enhanced transmissibility in humans can arise through
reassortment between human seasonal and animal influenza strains, including the generation of viruses of
HA subtypes that are “novel” to the human population (e.g. the 1957 H2N2 pandemic virus and the 1968
H3N2 pandemic virus).
Animal influenza viruses that continue to infect humans, in particular swine H3N2v viruses and avian
influenza H5N1, H7N9, and H9N2 viruses, do not efficiently infect or transmit in people. However, some
of these viruses share phenotypic properties of viruses that do efficiently transmit in humans, including
the ability to transmit via the respiratory route in ferrets and the ability to binding “human-like” sialic acid
receptors, and computational modeling suggests that the set of adaptive mutations needed to confer the
capacity for airborne transmission in mammals to H5N1 viruses can accrue during a single round of
transmission in a human host. The evolutionary implications of these findings – i.e. whether these viruses
are likely or unlikely to directly evolve the capacity for efficient transmission in humans – are

unknowable, given the small number of pandemics from which to draw lessons about the natural
evolution of human transmissibility. Critically, the fact that fully avian influenza strains have adapted to
efficiently transmit between dogs definitively demonstrates that cross-species adaptation of avian viruses
to mammals is possible. Furthermore, lessons learned from experiments that enhance the transmissibility
of fully avian or swine strains may be generalizable to mixed-species reassortant strains, thus their value
does not depend on whether the strains under study are likely to directly evolve enhanced transmissibility.
9.13.2 Introduction
Gain-of-function (GoF) experiments can be classified into two broad categories based on the purpose and
outcomes of the approach: (1) experiments that generate tools for scientific or public health use and (2)
experiments that enhance scientific understanding of virus behavior. The “tool” category of approaches
includes those that generate knowledge or products for use in vaccine production, such as high-yield
candidate vaccine viruses (CVVs) and knowledge about molecular markers that improve CVV growth
and those that adapt viruses for growth in mice or ferrets to generate animal models. These approaches are
not designed to generate or study phenotypes that are likely to occur in nature, and their benefits derive
solely from use of the information/tools for further scientific study or for MCM development/production.
The second category of GoF experiment generates scientific information that enhances the understanding
of virus physiology and behavior, which improves scientific knowledge and may additionally benefit
public health. This category includes GoF approaches that enhance the infectivity and transmissibility of
animal influenza viruses in mammals, that enhance the pathogenicity of influenza viruses in appropriate
animal models, that lead to evasion of existing natural or induced immunity, that lead to evasion of
therapeutics, and that involve reassortment between two different virus strains. Findings from these
approaches demonstrate what is possible for viral physiology and behavior in model systems and in a
laboratory environment. Importantly, the scientific relevance of this information and its utility for public
health depends on whether the phenotypes under study are likely to arise in nature. For example, using
information about molecular markers of mammalian adaptation in avian influenza viruses to prioritize
pandemic preparedness investments may be inappropriate if avian influenza viruses are unlikely to evolve
to efficiently infect humans in nature. As efforts to study these phenotypes aim to directly or indirectly aid
efforts to mitigate the public health consequences of seasonal influenza epidemics and influenza
pandemics, the likelihood of GoF phenotypes arising in nature also speaks to the risk of not pursuing GoF
research.
To provide context for our evaluation of the benefits of this research, this report will evaluate the
likelihood that the four GoF phenotypes listed in the paragraph above will arise in nature. Within each
phenotypic category, we first briefly review relevant GoF studies and results. Next, we draw upon several
types of evidence to evaluate whether the phenotype is likely to arise in nature, namely characterization of
wildtype viruses, epidemiological studies, and computational modeling approaches.
9.13.3 Evasion of existing natural or induced immunity (Antigenic Drift)
GoF approaches in this phenotypic category experimentally induce antigenic drift of seasonal influenza
viruses in the laboratory through serial passage of viruses in the presence of cognate antibodies or through
targeted mutagenesis to introduce mutations expected to confer antigenic change. These approaches
provide insight into the mechanisms underlying antigenic drift and also generate information that may
benefit antigenic surveillance of seasonal influenza viruses and strain selection for seasonal flu vaccines.
Within the Framework definition of GoF, this phenotypic category includes experiments that generate
novel antigenicity-altering amino acid substitutions, which have not yet been observed in nature, as well
as those that test the phenotypic consequences of particular amino acid substitutions found in wild type

strains identified through surveillance. The likelihood that the phenotypic changes observed in the former
type of experiment, i.e. forcing antigenic drift of currently circulating influenza strains, is of interest for
this report.
Since the emergence of the seasonal H1N1 and H3N2 strains of influenza in human populations (i.e.
following the 1918 H1N1 pandemic and the 1968 H3N2 pandemic), both strains have drifted
antigenically in nature. For example, the H3N2 strain underwent ten antigenic changes (termed antigenic
cluster transitions) between its emergence in 1968 and 2004, typically drifting every two to four years. 769
The H1N1 strain has also drifted over time, exhibiting 16 antigenic changes between 1918 and 2008, with
each antigenic cluster circulating for one to ten years prior to drift.770 Antigenic variants of the 2009
H1N1 pandemic strain have been detected in nature but have not yet become widespread, such that the
H1N1 component of the seasonal flu vaccine has not changed since the emergence of the virus in
2009.771,772,773 These observations definitively demonstrate that antigenic drift of currently circulating
influenza viruses, as induced through GoF experiments, occurs regularly in nature.
9.13.4 Evasion of Therapeutics
GoF approaches in this phenotypic category experimentally generate antiviral-resistant strains through
serial passage of viruses in the presence of sub-inhibitory concentrations of therapeutic or through
targeted genetic modification to introduce mutations expected to confer antiviral resistance. These
approaches aim to gain insight into the mechanistic basis of antiviral resistance. An additional goal is the
identification of mutations that confer antiviral resistance for use in surveillance, which influences
therapeutic recommendations for seasonal influenza infections and pandemic preparedness initiatives for
animal influenza viruses. Within the Framework definition of GoF, this phenotypic category includes
experiments that confer antiviral resistance to particular strains that have not yet exhibited resistance in
nature as well as those that test the phenotypic consequences of mutations observed in wild type antiviral
resistant strains. As above, the likelihood that the phenotypic changes observed in the former type of
experiment will arise in nature is of interest for this report.
Mutations that confer resistance to both classes of licensed influenza antivirals, the adamantanes and the
neuraminidase inhibitors (NAIs) have arisen in nature. The adamantane class of antivirals, introduced into
clinical practice in the early 1960s, were widely used as the primary treatment for influenza for 40 years.
However, in the early 2000s, resistant strains emerged in nature, in particular strains carrying an S31N
mutation in the M2 protein, and quickly rose to worldwide prominence across multiple strain sub-types.
Specifically, the S31N mutation was identified in 0.4% of viruses in 1995 but its prevalence increased to
92% of viruses by 2006. 774,775 Widespread resistance persists, and the adamantanes are no longer
recommended for treatment.776
769 Smith DJ et al (2004) Mapping the Antigenic and Genetic Evolution of Influenza Virus. Science 305: 371-376
770 Liu M et al (2015a) Antigenic Patterns and Evolution of the Human Influenza A (H1N1) Virus. Sci Rep 5: 14171
771 Huang W et al (2015) Characteristics of oseltamivir-resistant influenza A (H1N1) pdm09 virus during the 2013-2014
influenza season in Mainland China. Virol J 12: 96
772 Makkoch J et al (2012) Whole Genome Characterization, Phylogenetic and Genome Signature Analysis of Human
Pandemic H1N1 Virus in Thailand, 2009–2012. PloS one 7: e51275
773 Ramos AP et al (2013b) Molecular and phylogenetic analysis of influenza A H1N1 pandemic viruses in Cuba, May 2009 to
August 2010. International journal of infectious diseases : IJID : official publication of the International Society for
Infectious Diseases 17: e565-567
774 Bright RA et al (2005) Incidence of adamantane resistance among influenza A (H3N2) viruses isolated worldwide from
1994 to 2005: a cause for concern. Lancet 366: 1175-1181
775 Bright RA et al (2006) Adamantane resistance among influenza A viruses isolated early during the 2005-2006 influenza
season in the United States. JAMA 295: 891-894

Although resistance to NAIs is not yet widespread, resistance to one or multiple NAIs has been observed
in wild type strains. Specifically, strains that are resistant to oseltamivir or zanamivir as well as strains
that are resistant to both drugs have been observed in nature, including human seasonal strains (i.e.
A/H1N1,777 A/H3N2,778 and B strains779) as well as animal influenza strains (e.g. H7N9).780 In fact,
resistance to oseltamivir in seasonal flu strains was widespread during the 2007 – 2008 and 2008 – 2009
seasons, and resistant strains continue to be sporadically detected.781,782 NAI resistance has been linked to
a variety of mutations, several of which were first discovered in the laboratory through GoF studies. For
example, a GoF experiment discovered that the combination of H274Y and E119D mutations (N1
numbering) conferred pan-resistance to all three licensed NAIs (oseltamivir, zanamivir, and peramivir). 783
This set of mutations was later found to arise in an immunocompromised individual subjected to multiple
NAI treatment regimens over a prolonged course of illness, with minimal effects on viral growth.784
Taken together, these observations definitively demonstrate that NAI resistance has evolved and is likely
to continue to evolve in nature, and that particular antiviral resistance mutations identified through GoF
studies have naturally arisen in human populations.
9.13.5 Enhanced Pathogenicity
GoF approaches in this phenotypic category experimentally generate more virulent viruses in
representative model systems through serial passage of viruses in cells or animals or through targeted
genetic modification to introduce traits expected to enhance virulence (including reassortment and
targeted mutagenesis). These approaches aim to identify genetic and phenotypic traits underlying
pathogenicity, which provides insight into basic virulence mechanisms and can inform pandemic risk
assessments of circulating animal influenza viruses. Additionally, an improved understanding of how
viruses cause disease provides a foundation for the development of new therapeutics, in particular
therapeutics that protect against the severe disease observed during infection with highly pathogenic avian
influenza (HPAI) viruses such as H5N1.
A wide range of virulence has been observed in influenza strains that have infected humans, as detailed in
Table 9.3.
780 Sleeman K et al (2013) R292K substitution and drug susceptibility of influenza A(H7N9) viruses. Emerging infectious
diseases 19: 1521-1524
781 Dharan NJ et al (2009) Infections with oseltamivir-resistant influenza A(H1N1) virus in the United States. JAMA 301: 1034-
1041
782 Hauge SH et al (2009) Oseltamivir-resistant influenza viruses A (H1N1), Norway, 2007-08. Emerg Infect Dis 15: 155-162
783 Baek YH et al (2015) Profiling and characterization of influenza virus N1 strains potentially resistant to multiple
neuraminidase inhibitors. Journal of virology 89: 287-299
784 L'Huillier AG et al (2015b) E119D Neuraminidase Mutation Conferring Pan-Resistance to Neuraminidase Inhibitors in an
A(H1N1)pdm09 Isolate From a Stem-Cell Transplant Recipient. J Infect Dis

Table 9.3. Viruses Display Different Levels of Virulence
Virus CFR
Pandemic H3N8 (“Russian Flu”, 1889) 0.1 – 0.28% 785
Pandemic H1N1 (“Spanish Flu”, 1918) 2-3% 786
Pandemic H2N2 (“Asian Flu”, 1957) ~0.1% 787,788
Pandemic H3N2 (“Hong Kong Flu”, 1968) ~0.1% 789,790
Pandemic H1N1 (2009) 0.4% 791,792
H5N1 outbreaks 53% 793
H7N9 outbreaks 40% 794
Seasonal strains 0.01 – 0.5%795
Notably, there is a 100- to 1000-fold difference in the estimated case fatality rate (CFR) for seasonal
influenza viruses versus the 1918 H1N1 pandemic strain. Although the difference in observed CFR may
be partly explained by poor public health knowledge and capabilities in 1918 relative to the modern era,
experimental studies in ferrets also demonstrate that the 1918 H1N1 strain is highly pathogenic relative to
modern H1N1 viruses. 796 Other pandemic strains (1957 H2N2, 1968 H3N2, and 2009 H1N1) have also
exhibited higher virulence than seasonal influenza strains, albeit to a lesser degree than the 1918 H1N1
virus. Furthermore, H5N1 and H7N9 avian influenza strains that sporadically infect humans cause severe,
disseminated disease, exhibiting distinct cell and tissue tropism than human seasonal viruses. How
virulence would change if these strains were to adapt to efficiently infect and transmit in humans is
unknown.
785 Valleron A-J et al (2010) Transmissibility and geographic spread of the 1889 influenza pandemic. PNAS 107: 8778-8781
786 "Report of the Review Committee on the Functioning of the International Health Regulations (2005) in relation to Pandemic
(H1N1) 2009," World Health Organization, accessed August 25, 2015,
http://apps.who.int/gb/ebwha/pdf_files/WHA64/A64_10-en.pdf
787 Li FC et al (2008) Finding the real case-fatality rate of H5N1 avian influenza. Journal of epidemiology and community
health 62: 555-559
789 Li FC et al (2008) Finding the real case-fatality rate of H5N1 avian influenza. Journal of epidemiology and community
health 62: 555-559
791 Vaillant L et al (2009) Epidemiology of fatal cases associated with pandemic H1N1 influenza 2009. Euro surveillance :
bulletin Europeen sur les maladies transmissibles = European communicable disease bulletin 14
792 Fraser C et al (2009) Pandemic Potential of a Strain of Influenza A (H1N1): Early Findings. Science 324: 1557-1561
793 WHO. Cumulative number of confirmed human cases for avian influenza A(H5N1)reported to WHO, 2003-2015.
Update November 13, 2015. Accessed November 28, 2015.
794 WHO. Influenza at the human-animal interface. Summary and assessment as of 17 July 2015.
http://www.who.int/influenza/human_animal_interface/Influenza_Summary_IRA_HA_interface_17_July_2015.pdf. Last
Update Accessed November 28, 2015.
795 Meltzer MI et al (2015) Standardizing scenarios to assess the need to respond to an influenza pandemic. Clinical infectious
diseases : an official publication of the Infectious Diseases Society of America 60 Suppl 1: S1-8
796 Bootsma MCJ, Ferguson NM (2007) The effect of public health measures on the 1918 influenza pandemic in U.S. cities.
PNAS 104: 7588-7593

Critically, it is not possible to predict the virulence and pathogenesis mechanisms of the next pandemic
influenza strain. However, the fact that past pandemic strains have exhibited higher levels of virulence
than seasonal strains, that 1918-like avian viruses are currently circulating in wild bird populations, and
that human infections with some H5 and H7 strains causes severe disease suggests that a virulent
pandemic strain could naturally emerge.797 This possibility lends support to the study of virulence using
GoF approaches, as these studies aim to generate knowledge that improves preparedness for pandemics
caused by highly virulent influenza strains.
9.13.6 Mammalian Adaptation and Enhanced Transmission in Representative Animal Models
GoF approaches in this phenotypic category experimentally generate viruses with enhanced infectivity
and transmissibility in representative animal models through serial passage of viruses in animals and/or
through targeted genetic modification to introduce traits expected to enhance infectivity or
transmissibility. These experiments aim to understand whether and how animal influenza viruses can
adapt to efficiently infect and transmit in humans, which provides insight into the mechanisms underlying
mammalian adaptation and transmissibility. This information also facilitates monitoring of the pandemic
risk posed by animal influenza viruses circulating in nature, which informs development of vaccines and
other pandemic preparedness initiatives that seek to mitigate the public health consequences of a
pandemic caused by animal-origin viruses. This phenotypic category includes experiments involving
animal influenza viruses (e.g. HPAI H5N1) as well as experiments involving reassortment viruses
comprised of gene segments from human seasonal and animal influenza viruses (e.g. an H5N1
reassortment strain comprised of an avian H5 gene and the remaining seven genes from the human
pandemic H1N1 strain).798,799 Experiments using both types of animal flu viruses have led to the
generation of modified viruses that are capable of transmitting between appropriate animal models
(guinea pigs, for contact transmission studies, or ferrets, for contact and airborne transmission studies).
Specifically, mammalian-transmissible variants of avian influenza H5N1 and H7N1 strains have been
generated in the laboratory, as well as mammalian-transmissible reassortment strains comprised of gene
segments from human seasonal viruses and either avian influenza H5N1 or H9N2 strains.800,801,
802,803,804,805,806
(Of note, serial passaging and/or reassortment studies involving other avian influenza
strains, such as serial passaging of H7N9 viruses, have not led to the generation of viruses with enhanced
transmissibility.)807
797 Watanabe T et al (2014a) Circulating avian influenza viruses closely related to the 1918 virus have pandemic potential. Cell
800 ibid.
802 Sutton TC et al (2014) Airborne transmission of highly pathogenic H7N1 influenza virus in ferrets. Journal of virology 88:
6623-6635
803 Wan H et al (2008) Replication and Transmission of H9N2 Influenza Viruses in Ferrets: Evaluation of Pandemic Potential.
PloS one 3
Li X et al (2014b) Genetics, receptor binding property, and transmissibility in mammals of naturally isolated H9N2 Avian
Influenza viruses. PLoS Pathog 10: e1004508
804 Chen L-M et al (2012) In vitro evolution of H5N1 avian influenza virus toward human-type receptor specificity. Virology
422: 105-113
805 Zhang Y et al (2013a) H5N1 Hybrid Viruses Bearing 2009/H1N1 Virus Genes Transmit in Guinea Pigs by Respiratory
Droplet. Science 340: 1459-1463
806 Sorrell EM et al (2009) Minimal molecular constraints for respiratory droplet transmission of an avian-human H9N2
influenza A virus. Proc Natl Acad Sci U S A 106: 7565-7570

Notably, experiments in this phenotypic category are distinct from others included in the Framework in
that avian or swine viruses that efficiently infect and transmit in humans have not yet evolved in nature.
While reassortant strains with genes of avian and/or swine origin have emerged to cause pandemics in
human populations, neither reassortant nor fully animal-origin strains of the H5, H7, or H9 sub-types,
which are thought to have the greatest pandemic potential of the avian influenza strains that have infected
humans, have evolved the capacity for efficient human transmission. Given the caveats associated with
translating laboratory results to nature, some have questioned whether the animal influenza strains used in
these GoF studies could ever naturally acquire enhanced infectivity and transmissibility in humans. As
mentioned above, the likelihood that this evolution could occur motivates the GoF studies and
qualitatively speaks to the risk of not investing in research that aims to mitigate the effects of future
pandemics caused by descendants of these viruses.
This section evaluates the likelihood that animal strains could evolve the capacity for efficient infection
and transmission in humans through either the direct evolution and/or the reassortment pathway. Three
types of evidence are reviewed: (1) epidemiological data about human infections with animal influenza
viruses, (2) laboratory data about the characterization of wild type viruses, detected through surveillance,
and (3) computational modeling of the capacity of wild type viruses to evolve mammalian
transmissibility.
9.13.6.1 Epidemiological Data
Relevant epidemiological data includes incidence, severity, and patterns of infection in humans (and non-
human mammals), as well as serological studies investigating population exposure to influenza viruses.
9.13.6.1.1 Cross-species adaptation events not involving humans
Although avian influenza (AI) viruses have not directly adapted to efficiently infect and transmit in
humans, AI viruses have directly evolved to efficiently transmit between other mammals. Namely, an
avian-origin H3N2 canine influenza virus emerged in dogs in the mid-2000s and is now circulating in dog
populations of China and South Korea, and possibly Thailand.808 Phylogenetic analysis revealed that
canine adaptation involved both intrasubtypic and heterosubtypic reassortment events as well as the
evolution of adaptive mutations. Isolated spillover events of avian influenza viruses in mammals have
also been detected, similar to humans. For example, in 2004, a dog was found to develop high fever and
lethargy following ingestion of duck carcasses. Necropsy revealed extensive H5N1 infection in the canine
tissues.809 In 2011, several New England harbour seals were found to be infected with an avian H3N8
virus that exhibited enhanced affinity for α2,6 receptors and was transmissible via respiratory droplets in
ferrets.810 Taken together, these examples demonstrate that avian influenza viruses have the capacity to
infect and evolve efficient transmissibility in non-human mammals.
9.13.6.1.2 Cross-species adaptation events involving humans
Numerous swine and avian influenza strains have infected humans, reviewed below. These data speak to
the current capacity for circulating zoonotic influenza strains to infect and transmit in people.
808 Zhu H et al (2015) Origins and Evolutionary Dynamics of H3N2 Canine Influenza Virus. Journal of virology 89: 5406-5418
809 Songserm T et al (2006) Fatal Avian Influenza A H5N1 in a Dog. Emerg Infect Dis 12: 1744-1747
810 Karlsson EA et al (2014) Respiratory transmission of an avian H3N8 influenza virus isolated from a harbour seal. Nat
Commun 5

Swine influenza strains H1N1v and H1N2v
Human infections with swine influenza strain H1N1v have been reported for decades, as far back as the
1930s.811 However, since 2005, only 19 cases of H1N1v infections in the U.S. have been reported to the
CDC, leading to one fatality, and human to human transmission has not been documented.812,813,814,815 Five
non-fatal cases of human infection with swine influenza strain H1N2v have also been reported. 816,817
Both variant viruses cause symptoms similar to seasonal strains. H1N1v infections have been reported in
several other countries, though in general surveillance for swine influenza infections is poor outside the
U.S. and Europe.818,819 Swine farm workers have been shown to have higher HI antibody titers against
H1N1 than the general population, suggesting that they are frequently exposed to H1N1 virus but
experience asymptomatic or sub-clinical infections.820
Swine influenza H3N2v
The first human case of infection with H3N2v was reported in the United States in July 2011, although
the virus was first detected in the U.S. stock of pigs in 2010. 821 As of 2015, 353 human infections with
H3N2v have been reported to the CDC, most of which occurred during outbreaks linked to agricultural
fairs in Ohio and Indiana in 2012.822,823,824 H3N2v illness is relatively mild; only 18 of the U.S. patients
were hospitalized and only one of those cases was fatal.825 Two clusters of cases – three children in Iowa
who visited the same health care provider and two children in West Virginia who attended the same day
811 Shope RE (1931) SWINE INFLUENZA : III. FILTRATION EXPERIMENTS AND ETIOLOGY. J Exp Med 54: 373-385
812 CDC. Reported Infections with Variant Influenza Viruses in the United States since 2005.
http://www.cdc.gov/flu/swineflu/variant-cases-us.htm#table-infections. Last Update September 4, 2015. Accessed
November 28, 2015.
813 Dacso CC et al (1984) Sporadic occurrence of zoonotic swine influenza virus infections. Journal of clinical microbiology
20: 833-835
814 Avian Flu Diary. http://afludiary.blogspot.com/2015/08/cdc-fluview-1-novel-h1n1v-case-reported.html. Last Update August
28, 2015. Accessed November 28, 2015.
815 Centers for Disease Control and Prevention. (2014a) Influenza Activity — United States, 2014-15 Season and Composition
816 ibid.
817 CDC. Reported Infections with Variant Influenza Viruses in the United States since 2005.
http://www.cdc.gov/flu/swineflu/variant-cases-us.htm#table-infections. Last Update September 4, 2015. Accessed
November 28, 2015.
818 Niemcewicz M et al (2013) Acute respiratory distress syndrome (ARDS) in the course of influenza A/H1N1v infection--
genetic aspects. Ann Agric Environ Med 20: 711-714
819 Calistri A et al (2011) Report of two cases of influenza virus A/H1N1v and B co-infection during the 2010/2011 epidemics
in the Italian Veneto Region. Virology Journal 8: 502
820 Olsen CW et al (2002) Serologic Evidence of H1 Swine Influenza Virus Infection in Swine Farm Residents and Employees.
821 Centers for Disease Control and Prevention. Seasonal Influenza (Flu): H3N2v and You.
http://www.cdc.gov/flu/swineflu/h3n2v-basics.htm. Last Update August 2014. Accessed September 2014.
822 “Reported Infections with Variant Influenza Viruses in the United States since 2005 | Swine/Variant Influenza (Flu),”
accessed August 26, 2015, http://www.cdc.gov/flu/swineflu/variant-cases-us.htm.
823 Greenbaum A et al (2015) Investigation of an Outbreak of Variant Influenza A(H3N2) Virus Infection Associated With an
Agricultural Fair-Ohio, August 2012. J Infect Dis
824 Centers for Disease Control and Prevention (CDC), “Notes from the Field: Outbreak of Influenza A (H3N2) Virus among
Persons and Swine at a County Fair--Indiana, July 2012,” MMWR. Morbidity and Mortality Weekly Report 61, no. 29 (July
27, 2012): 561.
825 CDC. Case Count: Detected U.S. Human Infections with H3N2v by State since August 2011.
http://www.cdc.gov/flu/swineflu/h3n2v-case-count.htm. Last Update September 4, 2015. Accessed November 28, 2015.

care and had no known contact with swine prior to symptom onset – suggest that H3N2v viruses are
capable of limited human-to-human transmission.826,827
Avian influenza H5Nx Strains
Highly pathogenic avian influenza H5N1 first caused human infections in 1997, following a poultry
outbreak in Hong Kong.828 Since 2003, 844 cases, 449 of which were fatal, were reported to the WHO,
representing a 53% case fatality rate.829 Most H5N1 cases have been in countries with a high prevalence
of backyard farming and active live poultry markets (LPMs), both providing opportunities for human
exposure to avian viruses through infected poultry.830 Several statistical models have attempted to
estimate the R0 of the H5N1 outbreaks, to determine whether the virus has the capacity for human-to-
human transmission; however, different research groups have generated drastically different estimates.
One group estimated an R0 value of 1.14, which meets criteria for self-sustaining transmission, but others
estimate the R0 of H5N1 closer to 0.2.831, 832 One major epidemiological case study in Vietnam gathered
strong evidence to suggest human-to-human transmission of H5N1, while other studies evaluating H5N1
infection patterns in family clusters suggested the converse. 833,834 Thus, the extent to which spillover
H5N1 viruses have any capacity for human-to-human transmission remains uncertain (and may vary by
strain). A recent seroepidemiological study in Egypt, a country with a large number of documented
human H5N1 cases, suggested that the prevalence of H5N1 infection is approximately 2% among
Egyptians exposed to poultry, though few of those exposed had experienced clinical symptoms of
infection.835 These data suggest that most H5N1 infections are asymptomatic or sub-clinical, such that the
“true” case fatality rate is much lower than that previously suggested based on the outcomes of the severe
cases that are reported to the WHO.
Only one other avain A/H5 strain has caused human infections – H5N6, which is also highly pathogenic
in poultry, has caused one fatal infection.836
Taken together, avian influenza H5N1 is capable of causing severe infections in humans, but
epidemiological and sero-epidemiological data suggests that the virus is poorly able to infect and transmit
in humans.
826 Centers for Disease C, Prevention (2011) Limited human-to-human transmission of novel influenza A (H3N2) virus--Iowa,
November 2011. MMWR Morb Mortal Wkly Rep 60: 1615-1617
827 Centers for Disease C, Prevention (2012) Update: Influenza A (H3N2)v transmission and guidelines - five states, 2011.
MMWR Morb Mortal Wkly Rep 60: 1741-1744
828 WHO. Avian Influenza Fact Sheet. http://www.who.int/mediacentre/factsheets/avian_influenza/en/. Last Update September
2014. Accessed November 28, 2015.
829 WHO. Cumulative number of confirmed human cases for avian influenza A(H5N1) reported to WHO, 2003-2015.
831 Ferguson NM et al (2004) Public Health Risk from the Avian H5N1 Influenza Epidemic. Science 304: 968-969
Aditama TY et al (2012) Avian influenza H5N1 transmission in households, Indonesia. PloS one 7: e29971
832 Yang Y et al (2007) Detecting human-to-human transmission of avian influenza A (H5N1). Emerging infectious diseases
13: 1348-1353
833 Tran TH et al (2004) Avian influenza A (H5N1) in 10 patients in Vietnam. N Engl J Med 350: 1179-1188
834 Olsen SJ et al (2005) Family Clustering of Avian Influenza A (H5N1). Emerging infectious diseases 11: 1799-1801
835 Gomaa MR et al (2015) Avian influenza A(H5N1) and A(H9N2) seroprevalence and risk factors for infection among
Egyptians: a prospective, controlled seroepidemiological study. J Infect Dis 211: 1399-1407
836 Pan M et al (2015) Human infection with a novel highly pathogenic avian influenza A (H5N6) virus: Virological and
clinical findings. J Infect

Avian influenza H7 Strains
Several subtypes of avian H7Nx have caused human infections: H7N2, H7N3, H7N7, and H7N9. Six
cases of H7N2 infection have been reported worldwide, most in patients who had been in close contact
with infected poultry prior to their infections.837,838 Although several patients were hospitalized, all
recovered from their infections. Several cases of H7N3 infection have also been documented in poultry
workers following contact with infected flocks; most experienced mild or sub-clinical infections, and all
patients recovered. 839,840 The first documented human H7N7 infection occurred in the UK in 1996, in a
woman who contracted the virus while cleaning her poultry shed. She exhibited mild symptoms and fully
recovered.841 The 2002-2003 human H7N7 outbreak in the Netherlands, which occurred as the result of
outbreaks in poultry populations, was the first non-H5N1 avian influenza outbreak in humans. Over 1,000
people had subclinical indications, 86 people were infected, including poultry workers and several of their
family members, and at least one person died from infection complications.842 H7N7 infections were
again documented in three poultry workers following a 2013 outbreak in Italy, all of whom displayed
mild symptoms and recovered. 843
As of November 2015, 683 people have been confirmed with a novel reassortant H7N9 virus, and 271
have died from the infection, representing a 40% case fatality rate.844 The majority of the infected are
elderly males with one or more underlying medical conditions.845 Persons infected with H7N9 often have
direct exposure to infected birds at live poultry markets. 846 Family cluster analysis has suggested limited
human-to-human transmission, but the restriction of transmission to within families hints at host-specific
susceptibilities to H7N9 infection.847 The R0 of H7N9 has been consistently calculated below one.
Specifically, the CDC calculated the R0 to be 0.06 during the first wave of infections and 0.35 during the
second wave, and other estimates have been similarly low.848,849 Serological analysis of Chinese poultry
workers revealed that 6% were seropositive for H7N9 infection but had experienced subclinical
indications of infection.850
837 Ostrowsky B et al (2012) Low pathogenic avian influenza A (H7N2) virus infection in immunocompromised adult, New
York, USA, 2003. Emerging infectious diseases 18: 1128-1131
838 Abdelwhab EM et al (2014) Prevalence and control of H7 avian influenza viruses in birds and humans. Epidemiol Infect
142: 896-920
839 Tweed SA et al (2004) Human illness from avian influenza H7N3, British Columbia. Emerging infectious diseases 10:
2196-2199
Lopez-Martinez I et al (2013b) Highly pathogenic avian influenza A(H7N3) virus in poultry workers, Mexico, 2012. Ibid.
19: 1531-1534
840 Puzelli S et al (2005) Serological analysis of serum samples from humans exposed to avian H7 influenza viruses in Italy
between 1999 and 2003. J Infect Dis 192: 1318-1322
841 Kurtz J et al (1996) Avian influenza virus isolated from a woman with conjunctivitis. Lancet 348: 901-902
842 Enserink M (2004) Infectious diseases. Bird flu infected 1000, Dutch researchers say. Science (New York, NY) 306: 590
843 Puzelli S et al (2014b) Human infection with highly pathogenic A(H7N7) avian influenza virus, Italy, 2013. Emerging
infectious diseases 20: 1745-1749
844 FAO. H7N9 Situation Update. http://www.fao.org/ag/againfo/programmes/en/empres/H7N9/Situation_update.html. Last
845 Watanabe T et al (2014b) Pandemic potential of avian influenza A (H7N9) viruses. Trends Microbiol 22: 623-631
846 Li Q et al (2014a) Epidemiology of Human Infections with Avian Influenza A(H7N9) Virus in China. New England Journal
of Medicine 370: 520-532
847 Jie Z et al (2013) Family outbreak of severe pneumonia induced by H7N9 infection. Am J Respir Crit Care Med 188: 114-
115
Qi X et al (2013) Probable person to person transmission of novel avian influenza A (H7N9) virus in Eastern China, 2013:
epidemiological investigation. BMJ 347: f4752
848 Kucharski AJ et al (2015) Transmission Potential of Influenza A(H7N9) Virus, China, 2013-2014. Emerging infectious
849 Chowell G et al (2013) Transmission potential of influenza A/H7N9, February to May 2013, China. BMC Medicine 11: 214
850 Yang S et al (2014) Avian-origin influenza A(H7N9) infection in influenza A(H7N9)-affected areas of China: a serological
study. J Infect Dis 209: 265-269

Taken together, H7N2, H7N3, and H7N7 have demonstrated limited capacities to infect humans and have
caused mild infections. In contrast, H7N9 has infected a large number of people over a short period of
time, relative to other avian influenza viruses, and causes severe infection. Similar to H5N1, sero-
epidemiological studies suggest that many H7N9 infections are asymptomatic or sub-clinical, so that the
“true” case fatality rate is likely lower than that estimated based on severe cases that interact with the
healthcare system.
Avian influenza H9Nx Strains
Since the first cases of human infection with avian influenza H9N2 in 1998 in Hong Kong, infections
have been sporadically reported in humans and have caused relatively mild infections.851,852,853,854,855
Epidemiological evidence suggests that H9N2 cannot transmit between people.856 A systematic review of
H9N2 seroprevalence in avian-exposed populations reported that between 1% and 43% of people had
evidence of H9N2 infection, a high level of exposure that suggests that many infections are sub-
clinical.857 Taken together, these data demonstrate that H9N2 has a limited capacity to cause mild
infections in humans and no current capacity for human-to-human transmission.
Avian influenza H10 Strains
Two H10Nx strains have infected humans: H10N7 and H10N8. An avian H10N7 outbreak occurred in
Australia during March of 2010. After culling, several abattoir workers displayed conjunctivitis and
minor respiratory distress, and H10 infection was confirmed in two workers.858 In December of 2013, an
elderly woman died of H10N8 that she acquired from a LPM in the Nanchang, China.859 Two subsequent
cases of H10N8 were identified in Nanchang, and one patient died.860 A serological analysis of H10N8
infection in LPM workers revealed that 21 had serological evidence of H10N8 infection despite no
clinical indications of viral infection.861 Taken together, these data demonstrate that H10Nx strains have
limited capacity to infect humans but may cause severe disease, and that these strains have no current
capacity for human-to-human transmission.
Reassortant strains
Human infections with reassortant strains containing avian H5, H7, or H9 genes, or the HA genes from
other avian and swine viruses that have caused human infections (listed above), have not been recorded.
However, all of the major influenza pandemics in the 20th and 21st centuries were caused by reassortant
851 Peiris M et al (1999a) Influenza A H9N2: aspects of laboratory diagnosis. Journal of clinical microbiology 37: 3426-3427
852 Peiris M et al (1999b) Human infection with influenza H9N2. Lancet 354: 916-917
853 Peiris M et al (1999c) Human infection with influenza H9N2. Lancet 354: 916-917
854 Butt KM et al (2005) Human infection with an avian H9N2 influenza A virus in Hong Kong in 2003. Journal of clinical
microbiology 43: 5760-5767
855 “WHO | Antigenic and Genetic Characteristics of Zoonotic Influenza Viruses and Candidate Vaccine Viruses Developed for
Potential Use in Human Vaccines,” WHO, accessed August 26, 2015,
http://www.who.int/influenza/vaccines/virus/characteristics_virus_vaccines/en/.
856 Uyeki TM et al (2002) Lack of evidence for human-to-human transmission of avian influenza A (H9N2) viruses in Hong
Kong, China 1999. Emerging infectious diseases 8: 154-159
857 Khan SU et al (2015) A Systematic Review and Meta-Analysis of the Seroprevalence of Influenza A(H9N2) Infection
Among Humans. J Infect Dis 212: 562-569
858 Arzey GG et al (2012) Influenza virus A (H10N7) in chickens and poultry abattoir workers, Australia. Emerging infectious
859 Chen H et al (2014) Clinical and epidemiological characteristics of a fatal case of avian influenza A H10N8 virus infection:
a descriptive study. Lancet 383: 714-721
860 Liu M et al (2015b) Genetic diversity of avian influenza A (H10N8) virus in live poultry markets and its association with
human infections in China. Sci Rep 5: 7632
861 Qi W et al (2014a) Antibodies against H10N8 avian influenza virus among animal workers in Guangdong Province before
November 30, 2013, when the first human H10N8 case was recognized. BMC medicine 12: 205

viruses that suddenly acquired the capacity for human to human transmission through antigenic shift. The
1918 H1N1 pandemic virus is thought to have arisen from reassortment between multiple avian strains,
and all subsequent pandemic strains (1957, 1968, and 2009) are reassortants comprised of human seasonal
and animal (avian and/or swine) gene segments.862,863,864,865,866 Specifically, the 1957 H2N2 pandemic
strain is a descendant of the 1918 H1N1 strain that acquired novel HA, NA, and PB1 genes from avian
viruses, the 1968 H3N2 strain is a descendant of the 1957 H2N2 strain that acquired novel HA and PB1
genes from avian viruses, and the 2009 H1N1 strain is a triple reassortant strain comprised of genes of
avian, swine, and human origin. Of note, the 1957 and 1968 pandemics were caused by HA subtypes that
were not previously known to readily infect and transmit in humans. Thus, the historical record
demonstrates that reassortment between human and animal viruses in nature can generate novel viruses
with enhanced transmissibility in people, including viruses of HA subtypes not previously associated with
human to human transmission. Of note, co-infection of people with H7N9 and either H3N2 or H1N1 has
been detected, which could provide opportunities for the generation of reassortant viruses with enhanced
transmissibility in people relative to the parental H7N9 strain.867,868
9.13.6.2 Laboratory Data – Characterization of wild type viruses
Isolates of swine and avian influenza from human infections have been characterized for properties
underlying mammalian adaptation and transmissibility, such as sialic acid receptor binding specificity, as
well as infectivity and transmissibility in representative animal models. Similar to epidemiological data,
these laboratory data speak to the current capacity for zoonotic influenza strains to infect and transmit in
mammals. Additionally, if a given virus does not efficiently infect or transmit in representative animal
models, the demonstration that it has acquired phenotypic properties thought to underlie mammalian
adaptation and transmissibility (e.g. the ability to bind α2,6 sialic acid receptors) may speak to its
potential to evolve the capacity for efficient infection and transmission of humans. That is, that virus may
be poised to adapt to more efficiently infect and transmit in humans. This section reviews the phenotypic
characteristics of wild type animal influenza strains isolated from human infections.
9.13.6.2.1 Swine influenza H1N1v and H1N2v
No studies have evaluated the sialic acid receptor binding specificity or the transmissibility of H1N1v or
H1N2v human isolates. Given that swine epithelial tissues express α2,6 sialylated receptors, it is likely
that both are capable of binding to α2,6 receptors.869
862 Smith GJD et al (2009c) Dating the emergence of pandemic influenza viruses. PNAS 106: 11709-11712
863 Antonovics J et al (2006) Molecular virology: was the 1918 flu avian in origin? Nature 440: E9;-discussion E9-10
864 Lu L et al (2014) Reassortment patterns of avian influenza virus internal segments among different subtypes. BMC Evol
Biol 14: 16
865 Kawaoka Y et al (1989) Avian-to-human transmission of the PB1 gene of influenza A viruses in the 1957 and 1968
pandemics. Journal of virology 63: 4603-4608
Scholtissek C et al (1978) Genetic relatedness between the new 1977 epidemic strains (H1N1) of influenza and human
influenza strains isolated between 1947 and 1957 (H1N1). Virology 89: 613-617
866 Smith GJD et al (2009a) Origins and evolutionary genomics of the 2009 swine-origin H1N1 influenza A epidemic. Nature
459: 1122-1125
868 Zhang W et al (2015) Co-infection with Avian (H7N9) and Pandemic (H1N1) 2009 Influenza Viruses, China. Emerging
869 Trebbien R et al (2011) Distribution of sialic acid receptors and influenza A virus of avian and swine origin in
experimentally infected pigs. Virology Journal 8: 434

9.13.6.2.2 Swine influenza H3N2v
Clinical isolates of H3N2v were shown to exhibit a preference for binding to α2,6 sialylated receptors, to
efficiently infect and transmit in ferrets by both contact and airborne routes of transmission, and to
efficiently replicate in human cell lines. Taken together, those observations suggest that H3N2v viruses
have the capacity for efficient replication and transmission in mammals.870
9.13.6.2.3 Avian influenza H5Nx Strains
Wild type isolates of H5N1 infect but do not transmit via the airborne route between ferrets.871 However,
viruses isolated from patients infected with H5N1 have demonstrated binding capability to both avian-like
α2,3 and human-like α2,6 receptors.872 Several other strains of H5Nx that have not caused human
infections have been evaluated for their virulence and transmissibility in ferrets as well as sialic acid
receptor binding specificity. Similar to H5N1 isolates, an H5N5 strain isolated from poultry has been
shown to bind both α2,3 and α2,6 sialic acids.873 The North American H5N2 and H5N8 viruses that
recently caused outbreaks in domestic poultry populations replicated efficiently in ferrets, but clinical
symptoms were mild and neither virus was able to transmit in a direct contact setting. 874 A European
H5N8 virus also exhibited low virulence in ferrets and was not transmitted via the respiratory route. 875
Multiple H7Nx sub-types that have infected humans have demonstrated the capacity to bind α2,6 sialic
acid receptors. Namely, an H7N2 virus isolated from poultry and patient isolates from the 2004 H7N3
outbreak in Canada exhibited enhanced affinity for α2,6 receptors, and H7N9 human isolates were
capable of binding both α2,3 and α2,6 receptors. 876,877,878 Multiple H7Nx strains have also been shown to
efficiently infect and transmit in ferrets. Human isolates from the Canadian H7N3 outbreak and an avian
H7N7 isolate were contact transmissible between ferrets, and a recent H7N9 human isolate had the ability
to transmit between ferrets via the airborne route.879,880,881
870 Pearce MB et al (2012) Pathogenesis and transmission of swine origin A(H3N2)v influenza viruses in ferrets. PNAS 109:
3944-3949
872 Shinya K et al (2005) Characterization of a Human H5N1 Influenza A Virus Isolated in 2003. Journal of virology 79: 9926-
9932
873 Li Q et al (2015) Novel reassortant H5N5 viruses bind to a human-type receptor as a factor in pandemic risk. Vet Microbiol
175: 356-361
874 Pulit-Penaloza JA et al (2015) Pathogenesis and Transmission of Novel Highly Pathogenic Avian Influenza H5N2 and
H5N8 Viruses in Ferrets and Mice. Journal of virology 89: 10286-10293
875 Richard M et al (2015) Low Virulence and Lack of Airborne Transmission of the Dutch Highly Pathogenic Avian Influenza
Virus H5N8 in Ferrets. PloS one 10: e0129827
876 Belser JA et al (2008) Contemporary North American influenza H7 viruses possess human receptor specificity: Implications
for virus transmissibility. PNAS 105: 7558-7563
877 Lopez-Martinez I et al (2013b) Highly pathogenic avian influenza A(H7N3) virus in poultry workers, Mexico, 2012.
Tweed SA et al (2004) Human illness from avian influenza H7N3, British Columbia. Ibid. 10: 2196-2199
878 Ramos I et al (2013a) H7N9 influenza viruses interact preferentially with α2,3-linked sialic acids and bind weakly to α2,6-
linked sialic acids. J Gen Virol 94: 2417-2423
Xiong X et al (2013) Receptor binding by an H7N9 influenza virus from humans. Nature 499: 496-499
879 Belser JA et al (2008) Contemporary North American influenza H7 viruses possess human receptor specificity: Implications
for virus transmissibility. PNAS 105: 7558-7563
880 Belser JA et al (2014) Influenza virus infectivity and virulence following ocular-only aerosol inoculation of ferrets. Journal
of virology 88: 9647-9654
881 Zhang Q et al (2013b) H7N9 influenza viruses are transmissible in ferrets by respiratory droplet. Science (New York, NY)
341: 410-414

9.13.6.2.5 Avian influenza H9N2 Strains
Characterization of H9N2 strains isolated from poultry in live poultry markets in China between 2009 and
2013 found that several exhibited a preference for binding to α2,6 sialic acid receptors (though retained
the ability to bind α2,3 receptor) and were capable of airborne transmission between ferrets.882
H10N8 viruses isolated from ducks have exhibited broad sialic acid receptor binding capabilities, to both
α2,3 and α2,6 receptors.883 H10N7 isolates from human and avian sources have also demonstrated broad
sialic acid receptor binding specificity. 884 Wild type isolates of neither strain have been characterized for
transmissibility.
9.13.6.3 Computational Modeling Data
Computational models for virus evolution can be used to explore the likelihood that a given set of
mutations shown to confer enhanced transmissibility in a laboratory setting will evolve in nature. For
example, following the identification of sets of mutations that were sufficient to confer airborne
transmissibility to H5N1 viruses by the Kawaoka and Fouchier research groups, another group evaluated
the likelihood that currently circulating H5N1 strains could evolve those mutations during passage
through a single human host.885 The authors consider several different evolutionary contexts including
various selection pressures, the need to acquire a different number of mutations (based on the number of
mutations in the starting virus), and varying lengths of infection time. The authors conclude that it is
possible for H5N1 to evolve the set of mutations shown to confer the capacity for respiratory droplet
transmission within a mammalian host, supporting the idea that the evolutionary pathway identified in the
laboratory studies is possible in nature.
Another research group used a modeling approach to predict the length of time needed for the H7 protein
from H7N9 viruses that have infected humans to acquire mutations that would render it structurally and
genetically similar to H3 proteins from human seasonal H3N2 viruses. Their model estimated that this
evolution, which may result in H7N9 viruses that are human to human transmissible, requires
approximately 11 years.886
Notably, the results of these and other evolutionary modeling studies are subject to significant uncertainty
due to uncertainties in the values of the parameters used to build the models, among other factors.
9.13.6.4 Conclusions
Laboratory experiments have enhanced the transmissibility of animal influenza strains that do not
efficiently transmit between humans in nature through the direct evolution pathway (i.e. the incorporation
882 Li X et al (2014b) Genetics, receptor binding property, and transmissibility in mammals of naturally isolated H9N2 Avian
Influenza viruses. PLoS Pathog 10: e1004508
Matrosovich MN et al (2001) H9N2 Influenza A Viruses from Poultry in Asia Have Human Virus-like Receptor Specificity.
Virology 281: 156-162
883 Deng G et al (2015) Genetics, Receptor Binding, and Virulence in Mice of H10N8 Influenza Viruses Isolated from Ducks
and Chickens in Live Poultry Markets in China. Journal of virology 89: 6506-6510
884 Ramos I et al (2015) Hemagglutinin Receptor Binding of a Human Isolate of Influenza A(H10N8) Virus. Emerging
885 Russell CA et al (2012) The Potential for Respiratory Droplet–Transmissible A/H5N1 Influenza Virus to Evolve in a
Mammalian Host. Science 336: 1541-1547
886 Peng J et al (2014) The origin of novel avian influenza A (H7N9) and mutation dynamics for its human-to-human
transmissible capacity. PloS one 9: e93094

of mutations through serial passaging or targeted mutagenesis) and/or through reassortment with seasonal
influenza viruses. To shed light on whether these laboratory-generated phenotypic changes could occur in
nature, three types of data were reviewed: epidemiological data about the number and patterns of human
infections with animal influenza viruses, laboratory data about the phenotypic characteristics of animal
influenza viruses isolated from human infections, and computational modeling data about the
evolutionary capacity of these viruses. The findings are summarized and synthesized below.
Avian and swine influenza viruses currently exhibit limited capacity to infect and transmit in humans,
though H5N1 and H7N9 viruses are capable of causing severe disease in the event of a human
infection.887,888,889,890 Human infections with reassortant viruses containing gene segments from avian or
swine viruses that have infected humans have not been observed, but co-infections of people with avian
and human seasonal viruses have been reported, which could provide opportunities for the emergence of
novel reassortant viruses with enhanced transmissibility in humans. Laboratory characterization of human
isolates of avian and swine flu viruses have shown that some H3N2v and H7N9 viruses are capable of
airborne transmission between ferrets. Other sub-types (including H5N1 and H9N2) do not transmit in
representative animal models, but human isolates of these viruses have the ability to bind “human-like”
α2,6 sialic acid receptors, thought to be critical for efficient infection and transmission in humans.
Collectively, these phenotypic data suggest that these viruses may have partially evolved the capacity for
human to human transmission. Finally, computational modeling suggests that the set of adaptive
mutations needed to confer the capacity for airborne transmission in mammals to H5N1 viruses can
accrue during a single round of transmission in a human host.
Taken together, the evolutionary implications of these observations – i.e. that some animal flu subtypes
(H3N2v, H5N1, H7N9, and H9N2) continue to infect humans and share some of the phenotypic
characteristics of viruses that do efficiently infect and transmit in humans – are uncertain. On the one
hand, fully avian or swine viruses are not known to have directly evolved the capacity for efficient
transmission in humans. Some have argued that the large number of human infections with these viruses,
including the many mild or sub-clinical infections that are indicated by sero-epidemiology studies, have
provided ample opportunities for transmissibility to evolve if that were possible. In particular, avian
influenza H5N1 strains first caused human infections over 15 years ago, in 1997. 891 On the other hand,
the historical record, comprising just four influenza pandemics, represents a scant source of data from
which to draw conclusions about what evolutionary pathways are or are not possible, as well as the length
of time that is or is not “sufficient” for a particular evolutionary change to occur. Moreover, the historical
record shows that influenza pandemics have occurred on average every 25 years, with an interim
pandemic period of up to forty years (i.e. 1918 and 1957 pandemics) – longer than the length of time that
H5N1 strains have been sporadically infecting people. In addition, socio-cultural factors that critically
influence the evolution of influenza viruses in human populations, in particular the nature of human
interactions with animals and the environment, change over time. These changes will further compromise
the relevance of predictions about viral evolution based on historical data. Critically, the fact that fully
avian influenza strains have adapted to efficiently transmit between dogs definitively demonstrates that
cross-species adaptation of avian viruses to mammals is possible.
What is clear from the historical record is that enhanced transmissibility in humans can arise through
reassortment between human seasonal and animal influenza strains, including the generation of viruses of
887 Gao R et al (2013) Human infection with a novel avian-origin influenza A (H7N9) virus. N Engl J Med 368: 1888-1897
888 Watanabe T et al (2013) Characterization of H7N9 influenza A viruses isolated from humans. Nature 501: 551-555
889 Hatta M et al (2001) Molecular basis for high virulence of Hong Kong H5N1 influenza A viruses. Science 293: 1840-1842
890 Katz JM et al (2000) Molecular correlates of influenza A H5N1 virus pathogenesis in mice. Journal of virology 74: 10807-
10810

HA subtypes that are “novel” to the human population (e.g. the 1957 H2N2 pandemic virus and the 1968
H3N2 pandemic virus). Importantly, lessons learned from laboratory studies focusing on fully avian or
swine strains, which explore pathways for directly evolving enhanced transmissibility, may be
generalizable to both wholly avian/swine influenza strains and mixed-species reassortment strains. For
example, both H5N1 transmissibility studies published in 2012 uncovered the same HA stability
phenotype underlying airborne transmissibility in ferrets, despite the fact that one study involved an HPAI
H5N1 strain whereas another involved a 7:1 reassortant with a seasonal H1N1 strain. Thus, even if avian
or swine strains are unlikely to directly evolve to efficiently transmit in humans in nature, transmission
studies involving fully avian or swine strains may provide information that is relevant to the behavior of
reassortment strains.
9.14 Evaluation of the Globalization Potential of GoF Research
9.14.1 Summary of Findings
Whether risks and benefits are equally distributed across populations is an important consideration in any
risk-benefit comparison. For GoF research involving PPPs, the risks are global. This section provides an
overview of the potential for select benefits of GoF research conducted in the U.S. to diffuse globally, in
order to inform the comparison of risks and benefits associated with this research. A fully referenced and
more thorough discussion of these benefits can be found in Appendix IV Sec. 15.9.
The potential for three types of GoF benefits to globalize are considered:
• Improvements to the production of egg- and cell-based influenza vaccines,
• Assistance in the development of new influenza and coronavirus small molecule antivirals, and
• Contributions to risk assessments of circulating animal influenza viruses (pre-pandemic), which

in turn inform prioritization of pandemic preparedness activities such as the development of pre-
pandemic vaccines.
9.14.1.1 Improvements to the production of egg- and cell-based influenza vaccines
Several developing countries have the capacity to directly harness GoF research that benefits the
production of egg- and cell-based influenza vaccines. Specifically, non-high income countries host 18
vaccine producers spanning eight countries, representing an increase in the number of producers and
vaccine-producing countries since 2010. However, the establishment of new influenza vaccine production
lines in foreign countries is a slow process – on the order of eight years or longer – and is hampered by
political, technical, and economic factors. Lack of demand for influenza vaccines in-country appears to be
a particularly important issue facing all producers, which is compounded by a lack of knowledge about
optimal vaccination strategies in tropical regions.
U.S. vaccine donations in the event of a pandemic provide a second pathway for GoF-derived benefits to
reach developing countries. The United States donated approximately 14% of the vaccines committed to
the WHO during the 2009 H1N1 pandemic response, which collectively were deployed to 77 countries.
However, in 2009 both vaccine donation and distribution were significantly delayed, and logistical
challenges associated with vaccine distribution further reduced and/or delayed the quantity of vaccine
doses that reached developing countries’ populations. Although some of these shortcomings have been
addressed in theory by the WHO Pandemic Influenza Preparedness Framework, the ability of the U.S. and

the WHO to provide donated vaccines in time to mitigate the effects of a high morbidity influenza
pandemic in the world’s developing countries remains untested.
9.14.1.2 Assistance in the development of novel influenza or coronavirus antivirals
The ability of foreign countries to establish production lines for new antivirals depends not only on their
technical and industrial capabilities but also on their ability to negotiate complex patent issues. In cases
where patent protections do not apply, the actual time needed to initiate commercial production of a U.S.-
designed or commercialized antiviral appears to be in the 1-5 year range. However, several companies in
developing countries rapidly activated production of influenza antivirals in less than six months in 2005-
2006, when their governments were preparing for a potential H5N1 pandemic, suggesting that a general
lack of demand for influenza antivirals appears to be keeping globalization in check.
The U.S. demonstrated its willingness to donate influenza antivirals during the 2009 H1N1 pandemic.
However, problems of timeliness of supply compounded issues of suboptimal use in-country. The WHO
Pandemic Influenza Preparedness Framework (developed in 2011) seeks to address timeliness issues but
remains untested.
9.14.1.3 Contributions to pandemic risk assessments of circulating influenza viruses
The demonstration that animal influenza viruses can acquire pandemic properties in a laboratory setting
may galvanize preparedness efforts in developing countries where the virus is circulating in agricultural
animal or wildlife populations.
Because most developing countries in which high-risk animal influenza viruses are circulating lack the
ability to assess the transmissibility and virulence of viruses in ferrets, data which critically inform
pandemic risk assessments, risk assessments are carried out in collaboration with WHO and laboratory
members of the GISRS (including the CDC). Similar to USG risk assessments, these risk assessments
incorporate information derived from GoF research, alongside epidemiologic and virologic data, and
environmental factors that influence the pandemic potential of the virus.
Downstream of a pandemic risk assessment, the ability of developing countries to implement prevention
and early detection measures in response to the detection of zoonotic influenza cases or outbreaks in
humans and/or animals varies widely, depending on the state of public health infrastructure, the
relationship between the Veterinary Services and Public Health sectors, and the resources for investing
the prevention and response activities. Although multiple developing countries in which zoonotic avian
influenza infections have been detected in human and/or bird populations within the past five years
currently have the capacity to produce pre-pandemic influenza vaccines in-country, 21 do not. As WHO
does not stockpile pre-pandemic vaccines, the lack of vaccine production capabilities in some at-risk
countries limits the globalization potential of GoF benefits related to pandemic risk assessments.
9.14.2 Introduction
risk-benefit comparison. For GoF research involving PPPs, the risks– that biosafety or biosecurity
To inform the assessment of global risks versus benefits, this section evaluates the globalization potential
of select GoF benefits. Specifically, the potential for the outputs of GoF research conducted in the U.S. to
benefit the health of human populations in low- and middle-income bracket countries, as defined by the

World Bank, is analyzed.892
Three types of GoF benefits are considered in this section:
• Benefits to the development and production of egg- and cell-based influenza vaccines,
• Benefits to the development of new antivirals for influenza viruses or coronaviruses, and
• Benefits to risk assessments of circulating animal influenza viruses (pre-pandemic), which may in
turn stimulate pandemic preparedness activities such as enhanced surveillance and the
development of pre-pandemic vaccines.
Currently, there are no FDA-approved vaccines for MERS-CoV or SARS-CoV.893,894 GoF research
involving CoV has potential to benefit the development of CoV vaccines, which is an active area of
research involving a variety of vaccine platforms. Which type of vaccine will prove to be most effective is
not yet clear based on current research. Because the resources and expertise that are required to develop
production capacity for different types of vaccines varies, the globalization potential and barriers to
globalization for hypothetical CoV vaccines cannot be evaluated. Similar uncertainties preclude
evaluation of GoF benefits to the development of new influenza vaccines. For these reasons, the
assessment of GoF benefits to vaccines is limited to those benefits to the development and production of
existing influenza vaccines.
The globalization potential of GoF benefits to therapeutics is evaluated based on case studies of the four
influenza antivirals that are currently licensed in developed countries, each of which is a small molecule
compound initially developed in a high-income country. This assessment assumes that setting up
hypothetical future production lines for new small molecule drugs targeting CoVs or influenza will
require a similar level of resources as was needed to set up production lines for existing influenza
antivirals. As a result, the conclusions herein about the globalization potential of GoF benefits to
therapeutics apply to GoF research involving both influenza viruses and CoVs that may inform the
development of new small molecule drugs.
As GoF research involving CoVs does not currently benefit surveillance or decision-making in public
health policy, the assessment of the globalization potential of GoF benefits to pandemic risk assessments
is limited to research involving influenza viruses.
Below, the globalization potential of each of the three GoF benefits list above is evaluated in turn.
9.14.3 Potential Benefit 1- Improvements in the design and production of vaccines
Several types of GoF research have potential to improve the development and production of egg- and cell-
based influenza vaccines, namely GoF research that enhances virus production, that leads to evasion of
therapeutics, that enhances pathogenicity, and that leads to evasion of existing natural or induced adaptive
immunity. In brief, GoF research that enhances virus production leads to the generation of higher-yield
892 This classification system is used by the World Health Organization.

The World Bank, “Country and Lending Groups,” http://data.worldbank.org/about/country-and-lending-groups. Accessed
July 7, 2015.
893 Centers for Disease Control and Prevention (CDC), “Middle East Respiratory Syndrome (MERS),” June 2, 2015,
http://www.cdc.gov/coronavirus/mers/about/prevention.html. Accessed July 7, 2015.
894 World Health Organization, “Severe Acute Respiratory Syndrome (SARS),” December 1, 2013,
http://www.who.int/immunization/topics/sars/en/. Accessed July 7, 2015.

vaccine viruses, which can improve the availability of pandemic flu vaccines and the efficacy of seasonal
flu vaccines by shortening vaccine production timelines. Increasing the yield of vaccine antigen per egg
or cell also reduces the manufacturing cost of the vaccine, which may translate to a lower cost per vaccine
dose. GoF research that enhances virulence and that leads to evasion of therapeutics may lead to the
identification of molecular markers for virulence and antiviral resistance, respectively, that can be
removed from vaccine viruses through targeted mutagenesis, thereby increasing the safety of the vaccine
production process. Finally, GoF research that leads to the evasion of existing natural or induced
immunity has potential to improve the strain selection process for seasonal flu vaccines, thereby
increasing their efficacy. Each of these benefits may be harnessed by developing countries through direct
application of GoF research outputs to indigenous influenza production lines, or may benefit developing
countries indirectly through U.S. seasonal and pandemic vaccine donations.
9.14.3.1 Capacity for direct application of GoF research outputs to foreign influenza vaccine
production
High yield candidate vaccine viruses (CVVs) for seasonal and pandemic influenza strains, which serve as
the basis for vaccine strains used for large-scale manufacturing of vaccines, are developed by WHO
Collaborating Centres (WHOCCs) for Influenza and other collaborating laboratories.895,896,897 The WHO
Pandemic Influenza Preparedness Framework stipulates that influenza CVVs be made available from
WHOCCs to any influenza vaccine manufacturer and any other laboratory who makes a request, as long
as the requestor meets appropriate biosafety requirements to receive the strain in question.898 The GISRS
provides the international framework for the sharing of such biological materials between laboratories
around the world.899 Thus, any GoF benefits to strain selection for seasonal flu vaccines (which
determines the composition of CVVs) are inherently global. Other GoF benefits to influenza vaccine
production, which involve the discovery of molecular markers for high yield, virulence, and antiviral
resistance, can be incorporated into vaccine viruses by CVV developers or vaccine manufacturers.
Therefore, developing countries with industrial capacity to produce influenza vaccines have the ability to
directly benefit from GoF research conducted in the U.S., through utilization of modified CVVs provided
by WHOCCs or through the application of GoF research findings to vaccine strains developed by
indigenous manufacturers. Altogether, the likelihood and timescale over which GoF benefits to vaccine
production can be realized depends on two factors: (1) for those countries that do not yet have influenza
vaccine production capabilities, the resources needed for the establishment of new influenza vaccine
production lines, and (2) for those countries that already have influenza vaccine production capabilities,
the country’s regulatory policies governing changes in vaccine strains. Although an assessment of
country-specific regulatory policies as they pertain to the use of modified vaccine strains is outside the
scope of the current study, the FDA does not require regulatory approval for the commercial use of
modified vaccine strains (i.e. there is no regulatory barrier for GoF benefits to vaccine production in the
U.S.).
895 World Health Organization (WHO), “Influenza: Influenza vaccine viruses and reagents,”
http://www.who.int/influenza/vaccines/virus/en/. Accessed July 7, 2015.
896 World Health Organization (WHO), “Influenza: Virus Sharing,” http://www.who.int/influenza/pip/virus_sharing/en/.
Accessed July 7, 2015.
897 World Health Organization (WHO), Pandemic influenza preparedness Framework for the sharing of influenza viruses and
access to vaccines and other benefits (Geneva: World Health Organization Press, 2011), p. 16-17,
http://apps.who.int/iris/bitstream/10665/44796/1/9789241503082_eng.pdf. Accessed July 7, 2015.
898 Ibid.
899 World Health Organization (WHO), “Global Health Observatory (GHO) data: Global influenza virological surveillance,”
http://www.who.int/gho/epidemic_diseases/influenza/virological_surveillance/en/. Accessed July 7, 2015.

production
Global influenza production capacity was most recently comprehensively surveyed in 2010 by the WHO.
The WHO study identified 14 manufacturers in middle-income countries, collectively marketing at least
eleven vaccines and developing at least another eight vaccines.900,901 No updated list of active human
influenza vaccine manufacturers in 2014 or 2015 has been made publicly available. A dataset of current
influenza producers was therefore compiled to compare the current influenza production situation with
that surveyed in 2010.902 The results are summarized in the figure below, and a reference list is provided
in Sec. 16.9.6.
Figure 9.6. Developing countries that host at least one company with an influenza vaccine currently on the
market are shaded in deep blue. Developing countries that host at least one company with R&D efforts for
the production of an influenza vaccine are shaded in light blue.
Analysis of the assembled dataset reveals that the number of active producers outside of high-income
countries has increased since 2010. In total, 18 companies in middle-income countries were found to be
actively producing influenza vaccines and at least 13 additional companies have R&D work for influenza
vaccines at various stages of completion, compared to 14 manufacturers with current or planned flu
900 Ibid.
901 The survey identified the following five middle-income countries as having domestic influenza vaccines: China, India,
Thailand, Indonesia, and Romania. Planned production lines were identified in the following nine middle-income countries:
Brazil, Egypt, Kazakhstan, Mexico, Serbia, South Africa, Thailand, Iran, and Vietnam.
902 This dataset was compiled from lists of vaccine manufacturers in the 2010 WHO survey, the Developing Countries Vaccine
Manufacturers Network (DCVMN) directories from 2014 and 2015, the International Federation of Pharmaceutical
Manufacturers & Associations’ Influenza Vaccine Supply Members list, and the U.S. Department of Health and Human
Services’ Influenza Vaccine International Capacity Building Portfolio, supplemented by searches of additional
manufacturers mentioned in the literature or in news reports.

vaccine production lines in 2010.903 However, as many of the new influenza vaccine manufacturers since
2010 are located in countries that already had influenza vaccine production capabilities, overall the
geographic distribution of production capacities outside of high-income countries has only moderately
expanded. Eight countries now produce influenza vaccines (up from five). Based on current R&D efforts,
an additional five countries may become influenza vaccine producers in the future.904
A lack of end-user demand appears to be a recurring and common problem that is preventing several of
the middle-income firms mentioned in this section from initiating or maintaining influenza vaccine
production. With respect to pandemic influenza vaccines, this issue stems from a lack of government
support to purchase vaccines for pandemic preparedness purposes. With respect to seasonal influenza
vaccines, this issue involves a lack of demand by individuals. Notably, the Chinese market experience has
demonstrated that domestic demand for seasonal influenza vaccine increases with the income level of
individuals, thus low domestic demand is to be expected outside of high income countries.905 This
demand issue is compounded by the fact that current recommendations for the strain composition of
seasonal influenza vaccines are geared toward countries in the Northern and Southern hemispheres with
well-defined flu seasons, such as the United States and Australia.906 In contrast, well-defined seasonality
does not always occur in tropical regions of the world; instead, low levels of influenza virus circulate
throughout the year. In these regions, optimal vaccination strategies, including whether Northern or
Southern hemisphere vaccines are more protective and when during the year vaccines are best deployed,
are not well understood. Research to better understand patterns of influenza transmission and seasonality
in the tropics, as well as how best to mitigate the public health burden associated with influenza through
vaccination, is ongoing. This research provides an important foundation for developing countries’ efforts
to bolster their vaccine production capabilities and increase in-country demand in the future.
Several U.S. programs seek to support the aforementioned ability of developing countries to produce
vaccines. Since seasonal vaccine production lines are adapted to produce pandemic vaccines, these
pandemic preparedness programs complement seasonal influenza production assistance, and vice versa.907
The U.S. HHS supports production capabilities abroad for seasonal and pandemic influenza vaccine
through funding provided by its Biomedical Advance Research and Development Authority (BARDA)
branch.908 Overall, BARDA has provided over $70 million in financial support to 13 companies in 12
middle-income countries seeking to develop influenza vaccine production lines since 2006.909,910, 911 Of
903 This count excludes companies based in Taiwan, as the World Bank classes “Taiwan, China” as a “high-income” economy,
separately from “China,” which it classes as an “upper-middle-income” economy. See:
July 7, 2015.
904 Namely: Egypt, Iran, Serbia, Thailand, and Vietnam.
905 Eliza Yibing Zhou, “Vaccine Development in China,” BioPharm International 20, no. 4 (April 2007): p.1,
http://www.biopharminternational.com/china-today-vaccine-development-china. Accessed October 29, 2015.
907 For U.S. context, see: Executive Office of the President, President’s Council of Advisors on Science and Technology,
[U.S.A.] “Report to the President on Reengineering the Influenza Vaccine Production Enterprise to Meet the Challenges of
Pandemic Influenza,” August 2010, https://www.whitehouse.gov/sites/default/files/microsites/ostp/PCAST-Influenza-
Vaccinology-Report.pdf. Accessed July 7, 2015.
908 PATH, “PATH’s Work in Vaccine Development: Low-cost influenza vaccine production,”
http://sites.path.org/vaccinedevelopment/influenza/vaccine-production-in-the-developing-world/. Accessed August 3, 2015.
909 These companies are: Acera de Birmex (Mexico), BCHT (China), BioFarma (Indonesia), Cantacuzino Institute (Romania),
GPO (Thailand), Instituto Butantan (Brazil), IVAC (Vietnam), RIBSP (Kazakhstan), Serum Institute of India (India), The
BioVac Institute (South Africa), Torlak Institute (Serbia), VABIOTECH (Vietnam), and VACSERA (Egypt).
910 U.S. Department of Health & Human Services, “International Influenza Vaccine Capacity Building Portfolio.”
911 United States of America, “Report on USA implementation of Article X of the Biological and Toxin Weapons Convention,”
Meeting of the States Parties to the Convention on the Prohibition of the Development, Production and Stockpiling of

the 13 companies that received support from BARDA, six appear to remain in the R&D phase, one has
ceased production of vaccines, one appears to have halted R&D efforts, and five currently produce
influenza vaccines. Impediments to the establishment of production lines include human factors (e.g.
alleged corruption delaying construction of manufacturing facilities), technical factors (e.g. contamination
of vaccine doses), and economic factors (e.g. lack of domestic demand). (For additional details, see Table
16.40 in Sec. 16.9.3). Thus, more than eight years after BARDA began its assistance program, roughly
two thirds of the funding recipients appear to lack an influenza vaccine product on the market. The four
successful companies demonstrate that some developing countries are able to develop, produce, and
market a new influenza vaccine given eight years. However, the human, technical, and economic
problems encountered by the other companies drive home the point that setting up new influenza vaccine
production lines is time-consuming and is a high-risk endeavor from a business perspective.
9.14.3.3 Capacity of GoF benefits to vaccine production to globalize through U.S. vaccine donations
The United States supports foreign seasonal and pandemic influenza vaccine stockpiles through direct
vaccine donations, which represents a different pathway for the globalization of GoF benefits related to
vaccine development and production. Specifically, any GoF-derived improvements to U.S. vaccine
development and production will indirectly benefit developed countries that receive U.S.-produced
vaccines through assistance and emergency response programs.
9.14.3.3.1 U.S. seasonal vaccine donations
The U.S. HHS’ Centers for Disease Control has recently begun donating seasonal vaccines in an effort to
increase seasonal influenza vaccination in developing countries. The U.S. CDC organizes the donation of
seasonal influenza vaccines as part of the vaccine donation portion of the Partnership for Influenza
Vaccine Introduction.912 Since 2012, domestic companies involved in the production, distribution, and
sales of seasonal influenza vaccines have donated up to 375,000 doses of seasonal vaccine annually to
developing countries.913,914,915 However, several factors significantly limit the impact of this program.
First, donations are “based on [the] availability of excess vaccine supply” and are therefore unpredictable
and potentially limited.916 Second, WHO guidelines stipulate that the vaccine must be licensed for use in
the recipient country, which excludes many countries without domestic influenza vaccine production
capabilities and relevant regulatory infrastructure.917 Finally, the timing of U.S. seasonal vaccine
donations may not match the recipient country’s influenza season, further limiting the number of
countries that may benefit from the donated vaccine doses.918 Taken together, these limitations
significantly constrain the number of countries that can receive U.S. donations under these programs.
Bacteriological (Biological) and Toxin Weapons and on Their Destruction, Meeting of Experts, Geneva, Switzerland,
August 4-8, 2014, BWC/MSP/2014/MX/INF.5, p.4 para. 10. Accessed July 7, 2015.
912 The Task Force for Global Health, “Partnership for Influenza Vaccine Introduction,” <http://www.taskforce.org/our-
work/projects/partnership-influenza-vaccine-introduction>.
913 Joseph Bresee, CDC, “Global Action Plan for Influenza Vaccines – II: CDC’s Supportive Activities,” GAP-II Partners
Meeting, Dubai, United Arab Emirates, March 18, 2013,
<http://www.who.int/phi/Day1_9_Bresee_GAP2_CDC_PM_Dubai2013.pdf>.
914 Alan R. Hinman, “Partnership for Influenza Vaccine Introduction (PIVI),” Dubai, United Arab Emirates, March 25, 2014,
p.2, <http://www.who.int/phi/DAY1_08_Panel2_Hinman_Panel2_PIVI_PM_Dubai2014.pdf>.
915 Centers for Disease Control and Prevention (CDC), “Laos and Nicaragua Protect High-Risk Persons from Influenza, with
Help from Donor Coalition and CDC,” <http://www.cdc.gov/flu/international/highlight-high-risk.htm>.
916 Alan R. Hinman, “Partnership for Influenza Vaccine Introduction (PIVI),” p. 5.
917 Ibid.
918 Ibid.

9.14.3.3.2 U.S. vaccine donations in response to a pandemic
In the event of a pandemic, U.S. national policy calls for donations of vaccines to the WHO for
redistribution to developing countries. As a member state to the WHO Pandemic Influenza Preparedness
Framework, the U.S. is committed to supplying influenza vaccines to a WHO-maintained pandemic
benefit-sharing system, which would then redistribute vaccines to developing countries as necessary to
respond to a pandemic.919 Although the exact quantity to be contributed by each member state is not
specified, the document makes clear that the vaccine donations should be structured as a percentage of
vaccine production runs, to ensure timely supply.920
The following case study, on the U.S. vaccine donations in response to the 2009 H1N1 pandemic, show
how and to what extent U.S. vaccine donations can reach developing countries. The 2009 pandemic
preceded and motivated the formation of the WHO’s Pandemic Influenza Preparedness Framework in
2011. As such, although the actions taken by the U.S. during the pandemic remain instructive, certain
shortcomings in the international donation and response system have been addressed by the establishment
of a Framework.
During the H1N1 influenza pandemic, U.S. vaccine donations were organized in response to 17 bilateral
requests and a call for “global solidarity” from the WHO Director General.921 In September 2009, the
United States pledged up to 10% of its vaccine production runs to the WHO; eight other countries
subsequently made similar pledges.922 The U.S. H1N1 influenza response established a “10%” rule of
thumb, whereby 10% of vaccine production runs would be donated to the WHO for distribution to
developing countries in need of assistance. In total, the United States donated 16,860,100 doses of 2009
H1N1 influenza vaccine to the WHO for international distribution, which represented approximately 14%
of the vaccines committed to the WHO.923,924 Out of a total of 122,450,000 vaccine doses committed by
all states, the WHO distributed a total of 78,066,290 doses of vaccines to 77 countries.925
Overall, donation of vaccines to WHO suffered from severe timeliness issues. Vaccine production and
domestic supply difficulties in the U.S. (and other developed countries) in turn delayed vaccine donations
to the WHO.926,927 Advanced purchase agreements, whereby a given number of vaccines not yet produced
919 World Health Organization, Pandemic influenza preparedness Framework for the sharing of influenza viruses and access to
vaccines and other benefits (Geneva: World Health Organization Press, 2011), p. 15-16, 18.
920 Ibid.
921 “An HHS Retrospective on the 2009 H1N1 Influenza Pandemic to Advance All Hazards Preparedness,” p. 86,
http://www.phe.gov/Preparedness/mcm/h1n1-retrospective/Documents/h1n1-retrospective.pdf.
922 The eight countries were: Australia, Brazil, France, Italy, New Zealand, Norway, Switzerland, and the United Kingdom.
World Health Organization, “Report of the WHO Pandemic Influenza A(H1N1) Vaccine Deployment Initiative,” 2012, p. 4,
http://www.who.int/influenza_vaccines_plan/resources/h1n1_deployment_report.pdf.
923 United States of America, “Identifying and addressing barriers to the emergency sharing of international public health and
medical assistance,” Meeting of the States Parties to the Convention on the Prohibition of the Development, Production and
Stockpiling of Bacteriological (Biological) and Toxin Weapons and on Their Destruction, Meeting of Experts, Geneva,
Switzerland, August 12-16, 2013, BWC/MSP/2013/MX/WP.6, p. 2 para. 5.
925 The commitment of vaccines to the WHO involves a signed agreement, and therefore goes beyond a political pledge.
World Health Organization, “Final Pandemic (H1N1) 2009 Vaccine Deployment Update,” November 10, 2010,
http://www.who.int/csr/disease/swineflu/action/h1n1_vaccine_deployment_final_update_2010_11_10.pdf.
926 David P. Fidler, Kelly Lee, “Negotiating Equitable Access to Influenza Vaccines: Global Health Diplomacy and the
Controversies Surrounding Avian Influenza H5N1 and Pandemic Influenza H1N1,” PLoS Med 7, no. 5 (May 2010),
927 Supriya Kumar et al., “US Public Support for Vaccine Donation to Poorer Countries in the 2009 H1N1 Pandemic,” PLoS
One 7, no. 3 (2012), http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3295778/.

are purchased by a government from a private vaccine producer, compounded accessibility issues.928
Since the vaccines already belonged to a particular buyer, the private firm was unable to donate a portion
of the run to the WHO, regardless of a desire to do so.929 Other developed countries were reticent in
donating vaccines, and in a particularly severe pandemic whether promised doses would reach developing
countries in time to be effective is unclear.930 Several developed countries– such as France, Germany,
Switzerland, and the Netherlands– tried to sell excess vaccines instead of donating them.931,932 The WHO
Pandemic Influenza Preparedness Framework’s explicit clause on the provision of vaccines on a rolling
basis seeks to prevent this particular donation timeliness problem, but whether countries will comply with
the Framework during a severe pandemic remains untested.933 In addition to delays in the donation of
vaccine doses, the planning and execution of the donation and distribution of vaccine doses and ancillary
supplies was hampered by several logistical, regulatory, and political factors that further delayed and/or
reduced the quantity of vaccine doses distributed to recipient countries.
Taken together, these challenges highlight that, while US donation of vaccines is a viable pathway by
which GoF benefits to vaccine production may globalize, the time needed to orchestrate the logistics of
vaccine shipment and vaccination in-country will delay delivery of a vaccine to a developing country’s
population relative to a scenario in which that country is capable of indigenously producing and freely
distributing its own vaccine doses.
9.14.3.4 Summary – Globalization potential of GoF benefits to influenza vaccine production
GoF benefits to the production of influenza vaccines can be realized by developing countries in two ways:
(1) through the direct application of GoF research insights to production in-country, and (2) through the
receipt of U.S.-produced vaccines donated through assistance or emergency response programs.
With respect to indigenous production capabilities, both the total number of vaccine producers outside of
high-income countries (17) and the number of non-high income producing countries (7) has increased
since 2010. As WHOCCs provide ready access to candidate vaccine strains to all such producers, these
countries are currently capable of harnessing GoF research benefits to vaccine production. The total
number of producers outside of high-income countries is slated to increase by as many as an additional
six countries in the near future given current R&D efforts by over a dozen companies spanning eight
different countries. Analysis of the R&D timelines for foreign influenza vaccine manufacturers that
received BARDA funding support shows that bringing a new influenza vaccine to market may require up
to eight years, and that many efforts to develop new production lines fail due to political, technical, and
economic factors. Lack of demand for influenza vaccines in-country appears to be a particularly
important issue facing all producers, which is compounded by a lack of knowledge about optimal
vaccination strategies in tropical regions. Therefore, whether current R&D efforts for the establishment of
new production lines will come to fruition is uncertain, and the rate of continued development of new
production capabilities in the future cannot be ascertained.
928 Sam F. Halabi “Obstacles to pH1N1 Vaccine Availability: The Complex Contracting Relationship among Vaccine
Manufacturers, the World Health Organization, Donor and Beneficiary Governments,” The Public Health Response to 2009
H1N1: A Systems Perspective, eds. Michael A. Stoto, Melissa A. Hidgon (New York: Oxford University Press, 2015), p.
207.
929 Ibid.
930 David P. Fidler, Kelley Lee, “Negotiating Equitable Access to Influenza Vaccines: Global Health Diplomacy and the
Controversies Surrounding Avian Influenza H5N1 and Pandemic Influenza H1N1.”
931 Ibid.
932 “La France veut revendre ses vaccins contre la grippe A,” [France wants to sell its vaccines against influenza A] Le
Parisien, January 3, 2010, <http://www.leparisien.fr/societe/la-france-veut-revendre-ses-vaccins-contre-la-grippe-a-03-01-
2010-763246.php>.
vaccines and other benefits, p. 15-16, 18.

U.S. donations of pandemic or seasonal flu vaccines provide a second pathway for GoF-derived benefits
to reach developing countries. The U.S. experience during the 2009 H1N1 pandemic demonstrated that,
although the U.S. was committed to providing some 10% of its vaccine stocks to developing countries
through the WHO, the effectiveness of these donations suffered from serious timeliness issues. Although
the WHO Pandemic Influenza Preparedness Framework (developed in 2011) established guidelines for
vaccine donation during a pandemic in an effort to address these shortcomings, the ability of the U.S. and
pandemic in the world’s developing countries remains unverified. The U.S. CDC organizes the donation
of surplus seasonal influenza vaccines from vaccine manufacturers to developing countries, but several
factors significantly limit the impact of this program.
9.14.4 Potential Benefit 2- Assistance in the development of new influenza or coronavirus antivirals
Several types of GoF research have the potential to inform the development of new influenza or
coronavirus antivirals, namely GoF research that alters host tropism, that enhances pathogenicity, and that
leads to evasion of antivirals. First, GoF approaches that enhance the virulence of influenza viruses or
CoVs may lead to the identification of novel virulence factors that are good therapeutic targets, thereby
enabling the development of novel therapeutics. Second, animal-adapted influenza viruses and CoVs
developed using GoF approaches that alter host tropism are used for testing the safety and efficacy of
candidate therapeutics. Third, GoF approaches that lead to evasion of therapeutics generation information
that is recommended for inclusion in an Investigational New Drug (IND) application to the FDA, thereby
facilitating regulatory approval of new therapeutics. These benefits may be harnessed by developing
countries either through indigenous production of new antivirals, or through direct U.S. donations of
antivirals in the event of a pandemic.
9.14.4.1 Capacity for foreign production of GoF-derived new influenza antivirals
The process by which a pharmaceutical company abroad can proceed to produce an antiviral compound
discovered in the U.S. is complex. When a novel compound showing medical promise is developed into a
potential treatment by scientists working for a company, the company typically owns the rights to the
discovery as per the scientists’ contracts, and is then free to patent the potential treatment.934 Countries
that do not recognize U.S. patents are free to produce the drug provided that no additional bilateral or
multilateral trade agreement clauses prohibits this activity. (For example, Tamiflu, which was originally
discovered and patented by Gilead Sciences, is not patent protected in Thailand, the Philippines, and
Indonesia.) 935,936 For countries where a U.S. patent is legally valid or where a U.S. invention has been
patented in-country, domestic producers can either obtain a license or challenge the patent’s validity by
producing the compound without a license.937 In practice, firms are often reluctant to license production in
order to maintain production line exclusivity, and governmental and public pressure has played a role in
convincing U.S. firms to grant licenses to foreign companies. For example, Roche was for threatened by
several Congress representatives with a temporary abrogation of the Tamiflu patent when the firm was
unable to meet demand during the 2005 H5N1 pandemic preparedness period, after which the company
reached a number of sub-licensing agreements with other companies abroad to produce the compound.938
934 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,” CRS Report for Congress, August 16, 2007, p. 7,
retrieved at http://www.ipmall.info/hosted_resources/crs/RL33159_070816.pdf.
935 Ibid.
936 Roche, “Factsheet Tamiflu,” November 17, 2006, p.6, <http://www.roche.com/tamiflu_factsheet.pdf>.
937 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,”.
938 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,” p. 3-4.

Indeed, national patent law traditionally allows governments to cancel medication patents or to force the
licensing of the compounds in response to medical emergencies.939
Patents protect a product for a significant period of time. For example, the first U.S. patent covering
Tamiflu was filed in 1996 by Gilead Sciences, and the company is still fighting in court attempts to
produce generic oseltamivir medication by referencing its patent protections.940,941 Once associated
patents on a compound and its manufacturing expire, all competitors are allowed to produce the
compound as a generic medication.942
The following section assesses the ability of foreign countries to establish production lines for notional
influenza or CoV antivirals developed in the U.S., based on case studies involving existing influenza
antivirals. As highlighted by the discussion above, deriving benefits from such a U.S. discovery relies not
only on a foreign country’s capacity to establish a production line but also its ability to negotiate complex
patent issues. Of note, the conclusions herein are based in the current state of patent and licensing laws.
These laws may change as the result of growing public and governmental pressure for affordable
medication at the national level, which has stimulated comprehensive multinational trading negotiations
that would potentially make it easier for pharmaceutical companies to obtain patents.943
9.14.4.1.1 Capacity for novel influenza antiviral production abroad
To evaluate the capacity of developing countries to establish production lines for new antivirals, the
globalization of production capabilities for the existing influenza antivirals zanamivir, oseltamivir, and
peramivir (approved for use in the U.S.), as well as for laninamivir octanoate (approved for use in Japan)
are used as case studies to estimate the length of time needed to establish production of a new antiviral.
Of note, all four antivirals are small molecule compounds, and all were discovered in high-income
(developed) countries. The development timelines for each antiviral compound are summarized in Table
9.4.
939 Donald G. McNeil Jr., “Indian Company to Make Generic Version of Flu Drug Tamiflu,” The New York Times, October 14,
2005, <http://www.nytimes.com/2005/10/14/health/indian-company-to-make-generic-version-of-flu-drug-tamiflu.html>.
940 Kali Hays, “Gilead Sues Lupin Over Plans To Produce Generic Tamiflu,” Law 360, September 17, 2015,
<http://www.law360.com/articles/703920/gilead-sues-lupin-over-plans-to-produce-generic-tamiflu>.
941 U.S. Patent 5,763,483 A, “Carbocyclic Compounds,” Filed December 27, 1996, Published June 9, 1998,
<http://www.google.com/patents/US5763483>.
942 World Health Organization (WHO), “Generic Drugs,” <http://www.who.int/trade/glossary/story034/en/>.
943 “Hard pills to swallow,” The Economist, January 4, 2014, <http://www.economist.com/news/international/21592655-drug-
firms-have-new-medicines-and-patients-are-desperate-them-arguments-over>.

Table 9.4. Information on influenza antivirals
Generic Proprietary Brand Category Year Earliest FDA approval, any formulation
name manufacturer 944 name compound
published
Neuraminidase
Zanamivir GlaxoSmithKline Relenza 1993.945 July 1999.946
inhibitors
Neuraminidase
Oseltamivir Roche Tamiflu 1997.947 October 1999.948
inhibitors
Neuraminidase
Peramivir Biocryst Rapivab 2000.949 Emergency use in 2009, approved for use in December 2014. 950
inhibitors
Laninamivir Biota Pharmaceuticals Neuraminidase Currently not FDA-approved; approved for use in Japan against
Inavir 2009.951
octanoate and Daiichi Sankyo inhibitors Influenza A and B since 2010 and 2013, respectively. 952
944 [WHO] Technical Studies Under Resolution WHA63.1, Final Document, A/PIP/OEWG/3/2, p. 117;
“Biota Reports That Laninamivir Octanoate is Approved for the Prevention of Influenza in Japan,” Biota, December 20, 2013,
http://investors.biotapharma.com/releasedetail.cfm?releaseid=815483.
945 Mark Von Itzstein et al., “Rational Design of potent sialidase-based inhibitors of influenza virus replication,” Nature 363 (June 1993): p. 418-423,
http://www.nature.com/nature/journal/v363/n6428/abs/363418a0.html.
946
U.S. Food and Drug Administration, “FDA Approved Drug Products: Drug Details, RELENZA,”
http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Search.DrugDetails.
947 Kim C. U. et al., “Influenza neuraminidase inhibitors possessing a novel hydrophobic interaction in the enzyme active site: design, synthesis, and structural analysis of
carbocyclic sialic acid analogues with potent anti-influenza activity,” J. Am. Chem Soc. (January 1997): p. 681-690, <http://www.ncbi.nlm.nih.gov/pubmed/16526129>;
948 U.S. Food and Drug Administration, “FDA Approved Drug Products: Drug Details, TAMIFLU,”
http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Search.DrugDetails.
949 Babu Y.S. et al., “BCX-1812 (RWJ-270201): discovery of a novel, highly potent, orally active, and selective influenza neuraminidase inhibitor through structure-based drug
design,” Journal of Medical Chemistry 43, no. 19 (2000): p. 3482-3486.
950 U.S. Food and Drug Administration, “FDA approves Rapivab to treat flu infection,” FDA News Release, December 22, 2014,
http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm427755.htm.
951 Makoto Yamashita et al., “CS-8958, a Prodrug of the New Neuraminidase Inhibitor R-125489, Shows Long-Acting Anti-Influenza Virus Activity,” Antimicrobial Agents and
Chemotherapy 53, no. 1 (2009): p. 186-192.
952 Biota Pharmaceuticals, Inc., “Biota Provides Update on BARDA Contract for Laninamivir Octanoate,” May 8, 2014,
http://investors.biotapharma.com/releasedetail.cfm?releaseid=846423.

All four compounds have been produced by some middle-income developing countries. Since companies
mostly do not report on R&D efforts nor publicize the terms regarding technology transfers of
sublicenses, finding out the average length of time necessary to establish production capability for a given
degree of technology assistance is very difficult. Efforts to develop production capabilities in developing
countries can nevertheless be broadly grouped into three strategies: licensed activities coupled with
follow-on research, independent ventures, and exploratory research in advancing of licensing or the
expiration of patents. Some examples of companies in middle-income countries are given below for each
strategy to qualitatively illustrate the challenges and timescale associated with each approach, although
limited details are available for some cases.
Licensed activities coupled with follow-on research
In China, the Shanghai Pharmaceutical Group and HEC Pharm Co. are the two companies licensed to
supply the Chinese state with oseltamivir.953,954 Under a restriction imposed by Roche, the producers can
“only use it for pandemic purposes within China”; in practice, the firms were not allowed to sell the
compound commercially and had to furnish oseltamivir to the state at regulated prices.955 Shanghai
Pharmaceutical Group announced they could produce 200,000 doses in six months when they obtained
their licensing agreement in December 2005.956 The amount of R&D time invested by the firm prior to
December 2005 to establish this oseltamivir production capacity was not revealed, but the announcement
came some eight years after oseltamivir was identified as a potential MCM in the published literature
(1997).957
Also in China, the firm Nanjing Simcere Dongyuan Pharmaceutical Co. Ltd., a subsidiary of Simcere
Pharmaceutical Group, obtained a license to produce and sell zanamivir in September 2006.958,959
According to a Simcere spokesman, GlaxoSmithKline licensed the production of the drug but only
provided “limited technical support” in its synthesis.960,961,962 Thus, a pathway was developed in-country
through joint academic-industry research.963 The firm obtained approval from the Chinese national
regulator to manufacture and sell the compound in China in 2010, and the firm is currently selling the
compound.964
953 Kirby Chien, Devidutta Tripathy, “China, India drug firms say primed for swine flu,” Reuters, April 30, 2009,
http://uk.reuters.com/article/2009/04/30/us-flu-drugs-generic-idUKTRE53T0UL20090430.
954 “Roche licenses China firm to produce Tamiflu,” China Daily, December 12, 2005, p.1-2,
http://www.chinadaily.com.cn/english/doc/2005-12/12/content_502758.htm.
955 Roche opens Tamiflu to outside firms,” Swiss Info, December 12, 2005, http://www.swissinfo.ch/eng/roche-opens-tamiflu-
to-outside-firms/4900404.
956 Wang Xu, “Shanghai firm wins license for generic version of Tamiflu,” China Daily, December 13, 2005,
http://www.chinadaily.com.cn/english/cndy/2005-12/13/content_502775.htm.
957 Kim C. U. et al., “Influenza neuraminidase inhibitors possessing a novel hydrophobic interaction in the enzyme active site:
design, synthesis, and structural analysis of carbocyclic sialic acid analogues with potent anti-influenza activity,” J. Am.
Chem Soc. (January 1997): p. 681-690, http://www.ncbi.nlm.nih.gov/pubmed/16526129.
958 GlaxoSmithKline, “Agreement to increase availability of Zanamivir supply in Asia and Lease Developed Countries,” May
15, 2007, http://www.gsk-china.com/asp/News/client/newconten/515200791555.htm.
959 PR Newswire, “Simcere Receives SFDA Approval to Manufacture and Sell Zanamivir in China,” Bloomberg, February 11,
2010, http://www.bloomberg.com/apps/news?pid=21070001&sid=aRO5.9_34evg.
960 Ibid.
961 “Scientists develop ways producing anti-bird flu drug Zanamivir,” People’s Daily, February 6, 2009,
http://en.people.cn/90001/90781/90878/6587151.html.
962 EffacttPharm, “Research Progress,” July 10, 2015, <http://www.effecttpharm.com/yifang_e.html>.
963 Shanghai Institute of Materia Medica, Chinese Academy of Sciences, “The New Drug Certificate for Anti-H1N1 Flu
Medicine Zanamivir granted to SIMM,” March 17, 2010, http://english.simm.cas.cn/rp/201003/t20100317_51500.html.
964 Simcere, “Zanamivir,” http://www.simcere.com/english/products/detail.asp?gongs_id=59&leibieid=APIs.

Independent ventures
The Indian company Cipla publicly announced in October 2005, during the heightened H5N1 pandemic
preparedness period, that it would independently produce oseltamivir without entering into a commercial
agreement with Roche.965 In a subsequent interview, the company chair declared that the company had
begun researching oseltamivir production techniques in 2004.966 In India today, Cipla Ltd., Ranbaxy
Laboratories, Strides Arcolab, and Natco Pharma all have production capacity for oseltamivir without
having entered into an agreement with Roche.967,968,969,970,971
Thailand took advantage of the fact that Tamiflu had not been patent-protected in-country and has had
independent production capacity for the generic oseltamivir since 2006.972,973,974 The Governmental
Pharmaceutical Organization manufactured 200,000 tablets in early February 2006, following an
announcement that it would do so in December 2005.975
Independent exploratory research
A number of research groups in developing countries publish research on synthesis pathway optimization
for newly discovered antiviral compounds. The ultimate objective of this type of research may be to
prepare for in-country industrial production of the antiviral in question, although end-use intent cannot be
definitely predicted based on publications in the scientific literature.
The chemical compound peramivir (first published in 2000 and approved for emergency use in the US in
2009 and for general use in 2014) has already been synthesized in a novel process by a Chinese research
team, which achieved this result by March 2012 at the latest.976 Unlike earlier publications that described
known pathways to obtain peramivir that were funded through grants for basic research projects on new
965 “The Tamiflu Manufacturing Controversy: An Interview with Yusuf Hamied,” Multinational Monitor vol. 27, no. 2,
March/April 2006, http://www.multinationalmonitor.org/mm2006/032006/interview-hamied.html.
966 Ibid.
967 “Resistant strain of swine flu feared; virus killing thousands in India,” Japan Times, February 26, 2015,
http://www.japantimes.co.jp/news/2015/02/26/asia-pacific/science-health-asia-pacific/resistant-strain-of-swine-flu-feared-
virus-killing-thousands-in-india/#.VcjIdfnZViY.
968 “Swine flu: Hetero Healthcare increases Fluvir production by 400%,” The Economic Times, February 26, 2015,
http://articles.economictimes.indiatimes.com/2015-02-26/news/59541921_1_swine-fluvir-oseltamivir.
969 Khomba Singh, “Govt curbs sale of flu drug Zanamivir,” The Economic Times, August 29, 2009,
http://articles.economictimes.indiatimes.com/2009-08-29/news/28483297_1_swine-flu-drug-oseltamivir-zanamivir.
http://uk.reuters.com/article/2009/04/30/us-flu-drugs-generic-idUKTRE53T0UL20090430.
971 “Ranbaxy to supply oseltamivir capsules to US,” The Economic Times, October 21, 2007,
http://articles.economictimes.indiatimes.com/2007-10-21/news/28461984_1_capsules-domestic-sales-generic-version.
972 “Tamiflu- Oseltamivir Production,” News Medical, February 1, 2011, http://www.news-medical.net/health/Tamiflu-
Oseltamivir-Production.aspx.
973 Pennapa Hongthong, “Scientists produce generic Tamiflu,” The Nation, August 4, 2006,
http://www.nationmultimedia.com/2006/08/04/national/national_30010320.php.
974 Roche, “Factsheet Tamiflu,” November 17, 2006, p.6, http://www.roche.com/tamiflu_factsheet.pdf.
975 Emma Chanlett-Avery et al., “International Efforts to Control the Spread of the Avian Influenza (H5N1) Virus: Affected
Countries’ Responses,” CRS Report For Congress, August 24, 2006, p. 17,
http://fpc.state.gov/documents/organization/65759.pdf.
976 Fei Jia, Juan Hong, Ping-Hua Sun, Jian-Xin Chen, Wei-Min Chen, “Facile Synthesis of the Neuraminidase Inhibitor
Peramivir,” Synthetic Communications: An International Journal for Rapid Communication of Synthetic Organic Chemistry
43, no. 19 (2013): p. 2641-2647, http://www.tandfonline.com/doi/abs/10.1080/00397911.2012.729279.

drugs,977,978 the Chinese research team developed a new pathway designed for industrial production.
These results demonstrate that domestic production of the compound is well within China’s technical
capabilities. The peramivir case is one in which a novel synthetic pathway for a U.S. designed chemical
was rapidly developed abroad, indeed even before the compound was approved for general use in the U.S.
by the FDA. Similarly, in December 2014, a Chinese research team published a novel synthetic pathway
for the production of laninamivir octanoate.979
As demonstrated by the above accounts, indigenous production of all four licensed influenza antivirals
has been pursued in middle-income countries, to varying degrees and through a variety of mechanisms.
Namely, indigenous production lines for zanamivir and oseltamivir have been established in several
countries, and Chinese research groups have demonstrated the capability to efficiently synthesize
peramivir and laninamivir octanoate, presumably in preparation for the eventual development of
production lines in-country. Although the amount of R&D time invested by each of the companies and
research teams named above to achieve their production capability is unknown (i.e. when the company
began researching synthetic pathways and/or began setting up production facilities), conservative
estimates demonstrate that at least some middle-income countries achieved the capacity for full-scale
production of a given MCM less than ten years after the compound was initially published in the
literature. Notably, several companies rapidly activated large-scale production capabilities in less than six
months in 2005-2006 when their governments were preparing for a potential H5N1 pandemic. This
suggests that, as with influenza vaccines, a general lack of demand for influenza antivirals appears to be
keeping production line globalization in check. Based on these cases, the actual time needed to initiate
commercial production of an antiviral designed in a developed country appears to be in the 1-5 year
range.
Of note, barriers to the establishment of antiviral production lines are likely to vary between different
types of therapeutics (e.g. small molecule drugs versus monoclonal antibodies), though patenting and
licensing issues are likely to be the same for all types.
9.14.4.2 U.S. antiviral donations
GoF benefits to the development of novel antivirals may also globalize through U.S. donations of
antivirals to developing countries. Current U.S. government assistance to antiviral supply abroad are
primarily limited to plans for donations to the WHO for redistribution to developing countries in case of
an influenza pandemic. As a member state in the WHO Pandemic Influenza Preparedness Framework, the
United States government is committed to contributing influenza antivirals to the WHO-organized
Pandemic Influenza Preparedness Benefit Sharing System, which would redistribute MCMs to third
countries as part of a pandemic response as needed.980 U.S. private pharmaceutical companies can and
have donated antiviral treatments to the WHO and to countries dealing with local outbreaks independently
977 顾轶娜,林东海,“新型抗流感病毒神经氨酸酶抑制剂帕拉米韦研究进展,”中国生化药物杂志 30, no. 4 (2009): p.273-276

[GU Yi-na, LIN Dong-Hai, “Research progress on peramivir as a novel anti-influenza virus neuraminidase inhibitor,”
Chinese Journal of Biochemical Pharmaceutics 30 no. 4 (2009): p.273-276.].
978 贾飞, 陈良柱, 陈建新, 孙平华, 陈卫民,“帕拉米韦合成路线图解,”中国医药工业杂志 42 no. 12 (2011): p. 954-956. [JIA
Fei, CHEN Jianxin, SUN Pinghua, CHEN Weimin, “Graphical Synthetic Routes of Peramivir,” Chinese Journal of
Pharmaceuticals 42, no. 12 (2011): p. 954-956.].
979 Tian J. et al., “Organocatalytic and scalable synthesis of the anti-influenza drugs zanamivir, laninamivir, and CS-8958,”
Angewandte Chemie 126 (2014): p. 14105-14108.

from government contributions.981,982 However, these private companies are under no obligation to do so
in the future, and hence the effect of this potential GOF-derived benefits dissemination pathway cannot be
reliably assessed.
As there are no licensed therapeutics for coronaviruses in the U.S. or abroad, neither the U.S. nor the
WHO have formal policies or plans in place for the donation of (notional) therapeutics in the event of an
epidemic caused by a novel coronavirus.
The following case study reviews U.S. donations of antivirals to foreign countries during the 2009 H1N1
pandemic and identifies bottlenecks that may pose a barrier to the globalization of GoF benefits via this
pathway in the future. Although the creation of the WHO Pandemic Influenza Preparedness (PIP)
Framework in 2011 limits the extent to which this case study is predictive of the successes and challenges
of influenza antiviral donation efforts in the future given its plan for a joint pre-pandemic influenza
antivirals stockpile,983 similar challenges could be encountered in the event of ad hoc donation of CoV
therapeutics during a CoV epidemic.
During the 2009 H1N1 pandemic, the U.S. initially donated 400,000 antiviral treatment courses to
Mexico, followed by 420,000 courses of oseltamivir for the Pan American Health Organization
(PAHO).984 PAHO then provided stocks to countries throughout Latin America and the Caribbean.985
Although this demonstrates U.S. willingness to provide antiviral doses in the event of a pandemic, one
U.S. public health policy stakeholder stated that the global health security enterprise may not be as willing
to donate antivirals in the event of future pandemics due to the expense associated with storing and
deploying the drugs.986 The use of donated antivirals during the H1N1 pandemic in developing countries
was in general suboptimal, in part due to the low availability of the antiviral compounds in recipient
countries.987 In Asia for instance, an authoritative review article noted that “health practitioners were
reluctant to follow the recommendation of the empiric use of oseltamivir”: the practitioners did not wish
to use scarce doses on ostensibly mild cases of influenza, even when the patient was in a high-risk
group.988
In sum, although U.S. policy supports the donation of influenza antivirals in the event of a pandemic, the
relatively small number of doses donated in comparison to the global need in the event of a pandemic
means that developing countries would face shortages, which would in turn exacerbate poor usage in-
country.
981 David Reddy, “Responding to pandemic (H1N1) 2009 influenza: the role of oseltamivir,” J. Antimicrob. Chemother. 65
supplement 2 (April 2010): ii35-ii40, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2835510/pdf/dkq014.pdf>.
982 Roche, “Factsheet Tamiflu,” November 17, 2006, p.6, http://www.roche.com/tamiflu_factsheet.pdf.
access to vaccines and other benefits (Geneva: World Health Organization Press, 2011), p. 18,
http://apps.who.int/iris/bitstream/10665/44796/1/9789241503082_eng.pdf.
986 Interview with US government official involved in public health preparedness and response decision-making for influenza
outbreaks.
987 Dale Fisher et al. “Pandemic response lessons from influenza H1N1 2009 in Asia," Respirology 16 (2011): p. 879,
http://onlinelibrary.wiley.com/doi/10.1111/j.1440-1843.2011.02003.x/abstract.
988 Ibid.

GoF research has the potential to benefit the development of novel therapeutics for influenza viruses and
coronaviruses. The ability of developing countries to establish production lines for such new antivirals
depend not only on their manufacturing capabilities but also on their ability to negotiate the complex
patent issues surrounding the marketing of therapeutics. In cases where patent protections do not apply,
case studies of international production of licensed influenza antivirals suggest that the time needed to
initiate commercial production of a U.S.-designed antiviral is one to five years. Patent protections do not
apply when a patent is not recognized nationally or is abrogated during a medical emergency, or where
the compound can be sublicensed from the patent owner. Notably, several companies in developing
countries rapidly activated influenza antiviral production capabilities to produce hundreds of thousands of
doses in less than six months in 2005-2006, when their governments were preparing for a potential H5N1
pandemic. This capacity for rapid scale-up of production suggests that the actual time needed for
establishment of a new production line may be much less than five years. As with influenza vaccines, a
general lack of domestic demand for influenza antivirals appears to be keeping globalization of GoF
benefits related to the development of novel therapeutics in check.
The U.S. demonstrated its willingness to donate antivirals during the 2009 H1N1 pandemic. However,
problems of timeliness of supply compounded issues of suboptimal use in-country. The WHO Pandemic
Influenza Preparedness Framework (developed in 2011) addresses these shortcomings but remains
untested.
9.14.5 Potential Benefit 3- Benefits to pandemic preparedness planning
This section assesses the globalization of GoF benefits that inform pandemic preparedness planning,
which includes two benefits. First, the demonstration that avian influenza viruses can evolve the capacity
for more efficient transmission in mammals may, in and of itself, stimulate interest and investment in
pandemic preparedness initiatives. Second, molecular markers for phenotypic properties of concern (e.g.
virulence, transmissibility, mammalian adaptation, and antiviral resistance), which are discovered and
validated using GoF approaches, inform pandemic risk assessments that guide prioritization of resources
for pandemic preparedness activities. The first benefit derives from GoF research that enhances the
transmissibility of influenza viruses in mammals, and the second derives from GoF research that enhances
the infectivity or transmissibility of influenza viruses in mammals, that enhances the virulence of
influenza viruses, and that leads to evasion of influenza viruses from therapeutics.
The extent to which these GoF benefits will globalize depends on whether and how information derived
from GoF studies influences decision-making about pandemic preparedness activities in countries in
which high-risk animal influenza viruses are circulating, as well as whether these countries have the
ability to engage in pandemic preparedness initiatives.
9.14.5.1 Role of GoF Research in Pandemic Risk Assessments for Developing Countries
First, the role of GoF research in pandemic risk assessments conducted by developing countries is
assessed. Two types of GoF studies are considered: (1) ““proof of principle” demonstrations that
particular animal influenza viruses can acquire pandemic properties (e.g. transmissibility) in the
laboratory, and (2) studies that establish molecular markers for phenotypic properties of concern
(transmissibility, virulence, etc.).

Although “proof of principle” experiments that demonstrate that an avian virus (e.g. H5N1) can acquire
the capacity for more efficient transmission in mammals have had minimal impacts on USG initiatives
due to the already high investments in pandemic preparedness, these GoF results have relatively greater
impacts on preparedness efforts in developing countries. One international public health official stated
that the experimental demonstration that H5N1 could evolve the capacity for airborne transmission in
ferrets was of “great importance” in countries where H5N1 was circulating.989,990,991 In response, some
countries mounted communications campaigns to engage with the public, public health personnel, and
health care workers about the risks associated with H5N1, in an effort to bolster their surveillance
capabilities. Thus, to date, these GoF experiments primarily benefit global rather than domestic
populations.
Most developing countries in which animal influenza viruses of concern (e.g. H5N1) are circulating are
not capable of conducting ferret experiments to evaluate the transmissibility and virulence of viruses,
which contribute critical data to a pandemic risk assessment (see Sec. 9.6.3.3).992 As a result, those
countries carry out risk assessments in conjunction with the WHO (as well as the CDC and other
laboratories in the GISRS as needed).993 This collaborative relationship is codified in the WHO’s
Pandemic Influenza Preparedness Benefit Sharing System, which states that WHO will seek to ensure that
member states and the WHO Secretariat “provide pandemic surveillance and risk assessment and early
warning information and services to all countries.”994 These assessments are conducted with input from
the Ministries of Health in a country of interest.995 Similar to risk assessments conducted by the USG,
WHO risk assessments consider the presence of molecular markers of mammalian adaptation,
transmissibility, and virulence, alongside virological data and in the context of environmental factors that
play important roles in the emergence of pandemic viruses.
Ultimately, the ability of a developing country to derive benefits from risk assessments informed by GoF
research will depend on the ability of the country to engage in responsive pandemic preparedness
activities. These include enhanced surveillance, implementation of community-level risk mitigation
measures, and pre-pandemic vaccine development.996 The following sections assess the potential for
developing countries to put in place such “downstream” responses.
9.14.5.2 Capacity for responsive public health activities in developing countries
Responsive capabilities are primarily relevant in countries in which zoonotic influenza viruses (or
influenza viruses with zoonotic potential) are currently circulating. As seen on the map below (Figure
9.7), most countries in the world have detected cases of zoonotic avian influenza in humans or in birds
991 (2015g) Interview with international researcher or international public health official.
993 Ibid.
access to vaccines and other benefits, p.15.
996 C. Todd Davis et al., “Use of Highly Pathogenic Avian Influenza A(H5N1) Gain-Of-Function Studies for Molecular-Based
Surveillance and Pandemic Preparedness,” mBio 5, no. 6 (December 12, 2014) http://mbio.asm.org/content/5/6/e02431-
14.full.

within the last five years. Notably, a lack of detected cases may be due to poor detection and reporting
capabilities rather than the absence of avian influenza.997
Figure 9.7. Countries that reported a detected case of zoonotic influenza in humans or birds within the last
five years.998,999,1000,1001
Many countries with AI detections are developing (low- or middle-income) countries, in particular most
countries with repeated detections (i.e. multiple years) and sustained outbreaks in domestic poultry
populations. Public health responses to zoonotic influenza outbreaks in developing countries are
particularly challenging due to limited resources for carrying out response activities and because of the
need for a strong and coordinated veterinary service – public health system. The veterinary services of
most developing countries greatly suffer from weak human organizational factors compounded by
resource constraints.1002 The lack of effective communication strategies for behavioral interventions that
will reduce risks of disease spillover, e.g. at poultry farms, live bird markets, etc., was also highlighted by
influenza researchers and public health experts as a major challenge.1003 Convincing the public to comply
997 Tiaji Salaam-Blyther, “The 2009 Influenza Pandemic: U.S. Responses to Global Human Cases,” Congressional Research
Service, June 23, 2009, p. 11,
https://www.acs.org/content/dam/acsorg/policy/acsonthehill/globalchallengesdiscussions/swineflu/crs-r40588-us-
responses.pdf.
998 H5N1, H5N6, H6N1, H7N2, H7N3, H7N7, H7N9, H9N2, H10N7, H10N8.
999 World Health Organization (WHO), “Disease Outbreak News (DONs),” <http://www.who.int/csr/don/en/>.
1000 World Health Organization (WHO), “Monthly Risk Assessment Summary, Influenza at the Human-Animal Interface,”
http://www.who.int/influenza/human_animal_interface/HAI_Risk_Assessment/en/.
1001 Food and Agriculture Organization of the United States, “EMPRES-i Global Animal Disease Information System,”
http://empres-i.fao.org/eipws3g/.
1002 J. Weaver et al., “Initial assessment of strategic plans for improving the performance of Veterinary Services in developing
countries: a review of OIE PVS Gap Analysis reports,” Rev. sci. tech. Off. int. Epiz. 32, no. 2 (2012): p. 631-645.
1003 (2015g) Interview with international researcher or international public health official.

with disruptive measures is difficult, and one expert noted the value of GoF research results in
strengthening the evidence basis for recommendations.
These challenges are highlighted by Vietnam’s response to a series of H5N1 outbreaks in poultry in 2004
– 2005, which led to multiple cases of human infection. Vietnam’s initially responded by eradicating
infected birds and implementing movement restrictions for poultry, which proved to be ineffective given
their lack of nationwide surveillance and coordinated response capabilities.1004 Vietnam then launched a
nationwide surveillance effort and institution a mass vaccination program for poultry. These measures
also met with limited success, due to problems with recognition and reporting systems, insufficient
collaboration between human and animal health sectors, a general lack of resources to implement “active
surveillance and research” and other factors. 1005,1006 Today, H5N1 is considered endemic in poultry in
Vietnam, and sporadic cases of human infection with H5N1 continue to be reported by Vietnam. 1007
In contrast, Thailand, which also experienced H5N1 outbreaks in poultry and human infections during
that same time period, was able to mount a robust public health response that eradicated the virus from
domestic poultry production systems.1008 The Thai government implemented enhanced surveillance for
human and poultry cases, coupled with aggressive measures to eradicate the virus from poultry
operations, including culling of infected birds, destruction of related productions (e.g. feed, bedding, etc.),
and poultry movement controls.1009, 1010 In addition, the government produced and sold oseltamivir tablets
at subsidized prices, starting with 200,000 tablets manufactured in February 2006.1011 As a result of these
response measures, the last reported human case of avian influenza in Thailand was in 2006 and the last
reported animal case of avian influenza was in 2008.1012,1013,1014
These case studies demonstrate the overarching importance of a strong public health sector in being able
to benefit from pandemic risk assessments through implementation of prevention activities. Importantly,
the example of Thailand highlights that a robust response to a significant public health risk in middle-
income countries is not impossible.
1004 Ricardo J. Soares Magalhaes, Dirk U. Pfeiffer, Joachim Otte, “Evaluating the control of HPAIV H5N1 in Vietnam: virus
transmission within infected flocks reported before and after vaccination,” BMC Vet Res. 6 (2010): p.1
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2898779/pdf/1746-6148-6-31.pdf.
1005 Xiu-Feng Wan et al., “Evolution of Highly Pathogenic H5N1 Avian Influenza Viruses in Vietnam between 2001 and 2007,”
PloSOne 3, no. 10 (October 2008): 1-12, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2565130/pdf/pone.0003462.pdf.
1006 Nguyen Tran Hien, “Avian Influenza In Vietnam: Situation and Lessons Learned,” p.17,
http://www.fao.org/docs/eims/upload/250718/aj167e00.pdf.
1007 Sharmi W. Thor et al., “Detection and Characterization of Clade 1 Reassortant H5N1 Viruses Isolated from Human Cases in
Vietnam during 2013,” PloS One 10, no. 8 (2015): p. 1-20,
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4526568/pdf/pone.0133867.pdf.
1009 Thanawat Tiensin et al., “Highly Pathogenic Avian Influenza H5N1, Thailand, 2004,” Emerging Infectious Diseases 11, no.
11 (November 2005), http://wwwnc.cdc.gov/eid/article/11/11/05-0608_article.
1011 Ibid.
1012 Ibid.
1013 OIE, World Animal Health Organization Database (WAHID), “Detailed Country(ies) disease incidence,”
http://www.oie.int/wahis_2/public/wahid.php/Diseaseinformation/statusdetail.
http://empres-i.fao.org/eipws3g/.

9.14.5.3 Capacity for pre-pandemic vaccine production
In addition to implementing community-level prevention and surveillance activities in response to a high-

risk pandemic risk assessment, developing countries could derive benefits from such assessments by
investing in pre-pandemic vaccine development and stockpiling. The influenza vaccine producers with
influenza vaccines on the market identified in developing countries (see Sec. 16.9.6) are all capable of
producing pandemic vaccine strains using CVVs obtained through the WHO framework, as explained in
Sec. 9.14.3.1 above. The map in Figure 9.8 shows an overlay of the developing countries with current
vaccine production capabilities and those in which zoonotic influenza viruses have been detected in bird
and/or human populations within the past five years. Only seven out of 28 developing countries with
zoonotic AI detections in humans or in bird populations over the past five years have the capacity to
produce vaccines in-country. This result highlights that a limited number of countries that may be at risk
of the emergence of a novel pandemic strain within their borders can benefit from pandemic risk
assessments through the development and stockpiling of pre-pandemic vaccines. Notably, WHO does not
stockpile pre-pandemic vaccines for use in developing countries, but is rather focused on ensuring real-
time access to pandemic vaccines during a pandemic as outlined in the Pandemic Influenza Preparedness
Framework.1015,1016
Figure 9.8. Overlay of low- and middle-income countries with current or planned influenza vaccine
production capabilities and those that have reported AI detections in birds to OIE within the past five years.
Regions with AI detections are outlined in red. Countries (or regions) without vaccine production capabilities
are shaded in white, countries with current vaccine production capabilities are shaded in dark blue, and
countries with planned vaccine production lines are shaded in cyan.
1015 Immunizations SWGoIVa. Influenza A (H5N1) Vaccine Stockpile and Inter-Pandemic Vaccine Use Background Document.
http://www.who.int/immunization/sage/meetings/2013/november/SAGE_WG_H5vaccine_background_paper_16Oct2013_v
4.pdf. Last Update Accessed October 31, 2015.

9.14.5.4 Summary – Globalization of GoF benefits that inform pandemic risk assessments
animal or wildlife populations. For example, the 2012 demonstration that H5N1 could evolve the capacity
for airborne transmission between ferrets triggered some developing countries to initiate communications
campaigns to raise awareness of the risks associated with H5N1 infections among the public, public
health personnel, and healthcare workers, in order to bolster early detection capabilities.
capabilities to conduct ferret experiments evaluating the transmissibility and virulence of viruses, data
which critically inform pandemic risk assessments, risk assessments are carried out in collaboration with
the WHO and laboratory members of the GISRS (including the CDC). Similar to USG risk assessments,
these risk assessments incorporate information derived from GoF research, alongside epidemiologic and
virologic data, and environmental factors that influence the pandemic potential of the virus.
the prevention and response activities. Thailand’s ability to eradicate H5N1 from their poultry production
system in response to widespread outbreaks in poultry populations as well as multiple human spillover
cases in 2003 – 2006 indicates that successful eradication campaigns are possible. However, the fact that
Vietnam continues to experience HPAI outbreaks since the initial 2004 – 2005 outbreak in the region
highlights the challenges for successfully carrying out response activities that mitigate the risk of avian
influenza spillover into human populations.
Although multiple developing countries in which zoonotic avian influenza infections have been detected
in human and/or bird populations within the past five years currently have the capacity to produce pre-
pandemic influenza vaccines in-country, 21 do not. As WHO does not stockpile pre-pandemic vaccines,
the lack of vaccine production capabilities in some at-risk countries limits the globalization potential of
GoF benefits related to pandemic risk assessments.

10 Potential Proliferation of GoF Research
Summary 446
Purpose and Approach 446
Methods 446
10.3.1 Definition 446
10.3.2 Key informant interviews 447
10.3.3 Publication analysis to determine interest 447
10.3.4 Identification and analysis of historical case studies of research proliferation 448
10.3.5 Analysis of funding data 448
Limitations 448
Findings 449
10.5.1 Interest in the research community 449
10.5.2 Availability of resources 450
10.5.3 Other Factors that Might Influence Proliferation 452
10.5.4 Case Studies 452
10.5.5 Proliferation Trends 454
10.5.6 Research Sites 461
10.5.7 NIH Funding 462
Conclusions 462

10.1 Summary
To understand the risk of any system, an important input is the frequency with which any step is taken. To
understand the risk of GoF research, therefore, one must estimate how many laboratories may be
performing GoF experiments. Given the prospective nature of the Risk-Benefit Assessment, we must
estimate the number of laboratories that may perform the work in the future should the moratorium be
lifted, which is termed research proliferation. To estimate the ability of GoF research to proliferate, we
estimate the size of the American research community with facilities suitable to perform the work,
sufficient funding and similar research interests and capabilities to GoF research. We supplemented this
analysis with a study of three case studies of research discoveries similar to GoF and how those
discoveries drove research in other laboratories. We identified a group of 40 active, well-funded
researchers in the US who have been performing, or have the capacity to perform, certain types of GOF
experiments involving influenza, MERS, and SARS viruses. We also found that this research is currently
not limited by access to containment facilities: in fact, the existence of at least several hundred BSL-3 or
higher level facilities offers potential for expansion. This maximum number is supported by our case
studies which showed that a new discovery in this field may proliferate to as few as one and as many as
70 new groups around the world within 10-15 years. As indicated by the authorship patterns, about half of
the new labs actively collaborate with laboratories that published earlier whereas about half involve labs
unconnected to these founders.
The goal of this task was to estimate the expansion potential of Gain of Function (GoF) research if the
United States Government (USG) funding pause is lifted. Simply put, we are trying to answer the
question of how many labs may be participating in GoF research in the next few years given the state of
the field today. This information is important for risk estimates, because the probability of most
laboratory incidents is proportional to the number of groups performing these experiments.
Research expansion, which we also call proliferation, depends on three factors:
1. Size of the interested and capable research community,

2. Availability of resources to conduct the research, and
3. To what extent and how quickly a discovery is taken up by the research community.
We aimed to quantify each of these factors. Interest in the research community was measured by the
number of laboratories that published GoF studies; availability of resources on funding levels and number
of BSL-3/4 facilities; and rate and extent of proliferation on three past discoveries in virology, chosen to
represent various aspects of GoF research.
10.3 Methods
10.3.1 Definition
This analysis was based on the types of the GoF research recommended for assessment by the National
Science Advisory Board for Biosecurity (NSABB),1017 as follows:
Advisory Board for Biosecurity. May 2015.
rch-APPROVED.pdf

• Pathogens included – seasonal influenza, highly pathogenic avian influenza virus H5N1, low
pathogenic avian influenza virus H7N9, Severe Acute Respiratory Syndrome (SARS), Middle
East Respiratory Syndrome (MERS), and
• Pathogen characteristics – enhanced pathogen production as a result of changes in the replication

cycle or growth, enhanced morbidity and mortality in appropriate animal models, enhanced
transmission in mammals, evasion of existing natural or induced immunity, resistance to drugs or
evasion of other medical countermeasures such as vaccines, therapeutics, diagnostics.
10.3.2 Key informant interviews
Interviews were conducted with 11 GoF researchers, with whom we discussed the following questions:
1. How many groups represent the GoF research community in the US?
2. How many BSL-3 and 4 facilities are currently available to do GoF research in the US?
3. In what way has the moratorium and the dialogue surrounding GoF research influenced your
trainees’ interest in doing this work in the future? and
4. Can you suggest a past discovery that would make a meaningful case study for GoF research?
Not all researchers had the same expertise and so did not answer all the questions.
10.3.3 Publication analysis to determine interest
To identify the size of the research community interested in GoF research, we used three methods. First,
we searched PubMed and Web of Science (WoS) databases to obtain citations to two studies discussed in
the scientific and policy literature as exemplary of GoF research.1018,1019,1020 Second, we leveraged the
similarity feature in PubMed to abstract all publications ranked as similar to these two articles.1021 Finally,
we queried PubMed and WoS with search terms that were derived from the NSABB Framework of GoF
research, such as “enhanced pathogenicity and influenza virus” and “enhanced transmissibility and SARS
virus” (full list of search terms is included in the Appendix I Sec. 12).1022 Articles in foreign languages,
reviews, book chapters, editorials, opinion pieces, and conference abstracts were excluded. The searches
were conducted in June – July 2015 and were limited to the past five years.
For each data set, we abstracted the names of last authors with three or more publications in order to
identify the most active groups. The resulting list was de-duplicated and compared to the names of
investigators who received notifications under the USG funding pause and the missing names were added
to make the final list of PIs.1023
1018 Web of Science. https://isiknowledge.com/

1019 Imai M, et al (2012) Experimental adaptation of an influenza H5 HA confers respiratory droplet transmission to a
reassortant H5 HA/H1N1 virus in ferrets. Nature 486(7403):420-8.
1020 Herfst S, et al (2012) Airborne transmission of influenza A/H5N1 virus between ferrets. Science: 336(6088):1534-41.
1021 The set of similar articles is generated by comparing words from the title, abstract, and MeSH terms using a word-weighted
algorithm. PubMed. http://www.ncbi.nlm.nih.gov/pubmed.
Advisory Board for Biosecurity. May 2015.
1023 Jocelyn Kaiser. 17 November 2014. Moratorium on risky virology studies leaves work at 14 institutions in limbo.
http://news.sciencemag.org/biology/2014/11/moratorium-risky-virology-studies-leaves-work-14-institutions-limbo

10.3.4 Identification and analysis of historical case studies of research proliferation
Our objective was to find three historical examples of discoveries that were made 10-15 years ago that
involved a virus, (ideally involving GOF studies with pandemic influenza, SARS, or MERS) and then
analyze to what extent such studies resulted in an expansion of related work. The choice of case studies
was informed by key informant interviews.
Once the paper describing the initial discovery was identified, we used the approach described in the
publication analysis section to construct data sets. Each publication on the final list was examined to
determine whether it included the following experimental approaches:
• SARS case study – use of animals infected with live virus,

• PB2 case study – use of animals infected with PB2 mutant influenza strains, and
• 1918 case study – use of animals infected with reconstructed 1918 influenza strain.
The articles that did not meet these criteria were excluded. The remaining papers were further examined
to exclude the studies that were performed in facilities with containment levels lower than BSL-2+ as
these studies were unlikely to be relevant to proliferation of GoF research because according to the CDC
guidelines, propagation of SARS, MERS, and pandemic influenza viruses in cell culture and their use in
the inoculation of animals requires BSL-3 containment facilities.1024,1025,1026 BSL-2+ was included based
on the assumption that certain types of experiments that satisfy our inclusion criteria (e.g. with seasonal
influenza) can be performed under this containment level. The resulting papers were analyzed for
publication year, author names, and author affiliations.
10.3.5 Analysis of funding data
We queried the federal funding database RePORTER using the terms “influenza,” “SARS,” and “MERS”
to obtain total funding levels and with the names of 40 PIs with interest in GoF research identified in the
earlier tasks to obtain individual-level data. Searches were limited to 2010-2014.1027
10.4 Limitations
Our study had several limitations. First, last authors were used to define a research group, which may not
always be accurate. Second, funding data included only NIH expenditures for research on influenza,
SARS, and MERS viruses. Third, we could not determine how much of this funding is spent on GoF
research because, unlike publications, publically available grant abstracts do not contain sufficient
information to make this determination. Third, it is not possible to establish causality between a
publication of the seminal paper and subsequent research efforts by other groups. Fourth, the
dendrograms represent only the links between the authors on papers included in each case study.
Therefore, they are a snapshot in time and are unlikely to show the full picture of proliferation. Finally,
depending on the nature of the discovery and many other factors in the research environment,
proliferation may take different paths than what emerged from our case studies.
1024 http://www.cdc.gov/sars/guidance/F-lab/app5.html
1025 http://www.cdc.gov/coronavirus/mers/guidelines-lab-biosafety.html
1026 http://www.cdc.gov/flu/avianflu/h7n9/risk-assessment.htm
1027 NIH Reporter database. http://projectreporter.nih.gov/reporter.cfm

10.5 Findings
10.5.1 Interest in the research community
To estimate the level of interest in conducting GoF research we performed bibliometric analysis. Papers
citing either of the two articles that generated influenza viruses transmissible by air in ferrets (Imai et al
and Herfst et al), papers rated as similar to these articles by PubMed, and hits to the queries presented in
Table 10.1 were used as sources of interested groups. These searches resulted in nearly 3000 papers for
influenza and 2000 for SARS and MERS (Table 10.1). Removing duplicate publications reduced the
number of entries to 1805 and 1558, respectively.
In order to identify active GoF groups, we abstracted the names of all authors with three or more
publications over a five-year period. The resulting sets contained 259 influenza and 102 SARS/MERS
authors, of which 35 were the last authors and assumed to be Principal Investigators/group leaders.
Authors based outside of the US were excluded. The list of 35 authors was compared against the names of
the researchers who received notifications from NIH under the moratorium, resulting in the addition of
five individuals, bringing the total number to 40 investigators (Table 10.2).1028 This list represents the
group that has research interests and skill sets that align well with GoF research.
Table 10.1. Search Results to Determine Level of Interest in GoF Research

Influenza SARS/MERS
Total publications 2738 1886
Total unique publications 1805 1558
Unique authors with 3+ pubs 259 102
Table 10.2. Names of Principal Investigators in Alphabetical Order

Baric, Ralph S Harrod, Kevin S Mehle, Andrew Subbarao, Kanta
Bouvier, Nicole M Harty, John T Morrey, John D* Taubenberger, Jeffery K
Compans, Richard W Heise, Mark T Orenstein, Walter A* Tompkins, Stephen M
Denison, Mark R Katze, Michael G Palese, Peter* Topham, David J
Enjuanes, Luis Kawaoka, Yoshihiro Pekosz, Andrew* Treanor, John J*
Feldmann, Heinrich Lenschow, Deborah J Perez, Daniel R Tripp, Ralph A
Frieman, Matthew B Lowen, Anice C Perlman, Stanley Tseng, Chien-Te K
Gallagher, Thomas M Manicassamy, Balaji Richt, Juergen A Tumpey, Terrence M
García-Sastre, Adolfo Martinez-Sobrido, Luis Schultz-Cherry, Stacey Webby, Richard J
Govorkova, Elena A Mccray, Paul B Steel, John Webster, Robert G
*Added based on the moratorium notifications
Authors based outside of the US were excluded
1028 Jocelyn Kaiser. Moratorium on risky virology studies leaves work at 14 institutions in limbo. Science Magazine.
http://news.sciencemag.org/biology/2014/11/moratorium-risky-virology-studies-leaves-work-14-institutions-limbo.

We also investigated the size of the GoF community via key informant interviews. Of the 11 researchers
interviewed, five provided estimates, which ranged from three to 20. One respondent said that it included
essentially all influenza researchers. Clearly, the estimate by the researchers themselves is colored by
their own internal definition of what GoF entails. These anecdotal data suggest that our estimate of 40
groups was several fold greater than the size predicted by the research community and probably represent
the upper bound of interested groups.
10.5.2 Availability of resources
10.5.2.1.1 NIH Funding
While the interest/skills in the research community are required to conduct GoF research, they are not
sufficient without funding support and appropriate containment facilities in which to conduct the
experiments. We used RePORTER database to obtain data on funding levels for influenza, SARS, and
MERS research. We found that between 2010 and 2014 the NIH expenditures ranged from approximately
$84M to $104M for SARS, from $0.6M to $12M for MERS, and from $654M to $816M for influenza
(Table 10.3). 1029,1030 Funding for SARS and influenza decreased and for MERS increased over the period
examined, which is consistent with the emerging status of MERS. The number of research projects
followed a similar pattern.1031
Table 10.3. NIH Funding for SARS, MERS, and Influenza Research
FY SARS MERS Influenza
Dollar Dollar Dollar
Number of Number of Number of
amount, amount, amount,
projects projects projects
million million million
2010 104 243 1.5 4 816 1095
2011 87 196 <1 3 678 954
2012 94 216 2 4 596 871
2013 110 185 5 8 677 857
2014 84 173 12 17 654 873
We also determined the level of funding from NIH for the 40 PIs who represent the GoF community
(names shown in Table 10.2). The data were collected over a five-year period to minimize year-to-year
variation and the mean for each funded PI was calculated. Figure 10.1 shows that all but four
investigators had NIH funding, with the amounts ranging from $250K to over $10M per year. The median
funding level for 36 PIs with funding was $1.5M per year. Interestingly, the data shows a bimodal
distribution with nearly all laboratories supported by grants that total drastically more or drastically less
than this median value. According to the NIH estimates, this amount exceeds the total for three average
R01 grants.1032 As a very rough estimate, each R01 can support three researchers and supplies for their
experiments. Consequently, based solely on the raw numbers sufficient funding is available to support
over 100 additional researchers or approximately 150 researchers in total. However, it is not possible to
1029 We found significant discrepancies in the funding data contained in RePORTER and FederalRePORTER. Because
RePORTER is a more established system, we used it as data source. RePORTER contains primarily NIH funding data.
1030 We found significant inconsistencies in the data in the Reporter and FederalReporter databases in funding amounts and the
number of projects. Reporter was ultimately used because it is a more established database.
1031 RePOPRTER outputs the number of projects, not grants. The number of projects is equal to or exceeds the number of grants.
1032 https://nexus.od.nih.gov/all/2014/01/10/fy2013-by-the-numbers/

accurately estimate how much of the available funding currently supports or can support GoF research
versus non-GoF research involving SARS, MERS, and influenza viruses.
14
Number of PIs 7
6
4 4
3
1 1
0
0 250K 500K 1M 1.5M 2M 5M 10M >10M

Funding per year
Figure 10.1: NIH Funding Levels for PIs with Research Similar to GoF.
10.5.2.1.2 Containment Facilities
According to the CDC guidelines, propagation of SARS, MERS, and pandemic influenza viruses in cell
culture and their use in the inoculation of animals requires BSL-3 containment facilities.1033,1034,1035
Experiments that do not involve the manipulation of pathogenic virus do not contribute to accident risk.
Consequently, availability of BSL-3/4 facilities limits the level of research activity that may contribute to
biosafety risk. To estimate the upper bound of this limit, we reviewed the literature to determine the
number of these facilities in the United States. Information on the BSL-4 facilities was available from
several sources, and the estimates ranged from five to eight (Table 10.4).
A completely reliable number of BSL-3 facilities was not found. One source put this number at 1,495 in
2010.1036 However, according to the GAO report, “no single federal agency has the mission to track and
determine the risk associated with the expansion of BSL-3 and BSL-4 labs in the United States, and no
single federal agency knows how many such labs there are in the United States.”1037 Importantly, all
estimates that we were able to find put the number of laboratories in the hundreds at a minimum, which
represents vastly more containment capacity than needed to support the 40 interested groups identified
above.1038,1039
1033 http://www.cdc.gov/sars/guidance/F-lab/app5.html
1034 http://www.cdc.gov/coronavirus/mers/guidelines-lab-biosafety.html
1035 http://www.cdc.gov/flu/avianflu/h7n9/risk-assessment.htm
1036 Jocelyn Kaiser (2011). Taking Stock of the Biodefense Boom. Science Vol. 333 (6047): 1214.
1037 High-Containment Biosafety Laboratories: Preliminary Observations on the Oversight of the Proliferation of BSL-3 and
BSL-4 Laboratories in the United States. GAO-08-108T. 2007.
1038 USA Today. http://www.usatoday.com/story/news/2015/05/28/biolabs-pathogens-location-incidents/26587505/
1039 High Containment Laboratories. National Strategy for Oversight is Needed. GAO-09-574. 2009.

Table 10.4. BSL-4 Facilities in the United States
O=operational
NO=not operational
E=expanding
GAO 2009 Kuhn, presentation FAS, website
NIAID Rocky Mountain Lab, Hamilton
O O O
MT
CDC, Atlanta GA O O O
Georgia State University, Atlanta GA O O O
DOD USAMRIID, Fort Detrick MD O O E
University of Texas Medical Branch,
O O O
Galveston TX
Southwest Foundation for Biomedical,
O O O
San Antonio TX
DHS National Bio and Agro-Defense
NO NO
Facility, Manhattan KS
Boston University NBL, Boston MA NO NO
DHS National Biodefense Analysis and
Countermeasures Center, Fort Detrick NO NO
MD
NIAID Integrated Research Facility,
NO NO
Fort Detrick MD
Virginia Division of Consolidated
NO O NO
Laboratory Services, Richmond VA
10.5.3 Other Factors that Might Influence Proliferation
Based on a small survey, Julie Pfeiffer recently suggested that the current debate about GoF research and
the moratorium are having a negative effect on the career choices of scientists in training. 1040 We explored
this phenomenon in key informant interviews with GoF researchers. The majority of respondents (7 out of
11 or 63%) confirmed that the moratorium has had a “chilling effect” on their trainees. While the
interviews were conducted at the laboratories affected by the funding pause, and those being interviewed
were probably unlikely to view the moratorium positively, the data indicate that future rates of
proliferation might be inhibited by workforce shortages due to uncertainty in the ability to conduct the
research, negative publicity, or other factors.
10.5.4 Case Studies
To assess how quickly a research discovery, once made, will propagate through the scientific community
and lead to additional labs conducting similar research, we used historical discoveries as case studies.
Please recall that we sought discoveries that were made in 2000-2005, if possible involved SARS, MERS,
or influenza virus, and could be reasonably expected to lead to GoF research.
1040 Pfeiffer JK (2015) Is the Debate and “Pause” on Experiments That Alter Pathogens with Pandemic Potential Influencing
Future Plans of Graduate Students and Postdoctoral Fellows? mBio 6(1): e02525-14.

The following discoveries were proposed by GoF researchers: the SARS animal model; growing
embryonic stem cells in culture; macaque animal model using chimeric HIV; the PB2-627K host
adaptation mutation; the CRISPR-Cas system; Influenza virus genetics; and reverse genetics for
coronavirus. Three of these discoveries best met all of our criteria (Table 10.5). Two of these, the SARS
animal model and the PB2-627K host adaptation mutation were included in the study. The third, reverse
genetics for coronavirus, was excluded because it was too similar to PB2 and another discovery,
reconstruction of the 1918 influenza virus (suggested by our own team), which was chosen because it also
met all of our criteria.
Table 10.5. Case Studies Suggested in Interviews

Discovery year 2000- Involve SARS, MERS, or Expected to lead to
2005 influenza virus GoF research
SARS animal model (2003) ✓ ✓ ✓
Growing embryonic stem cells
in culture (1998)
HIV chimeric virus macaque
animal model (1998) ✓
PB2-627K host adaptation

mutation in influenza virus ✓ ✓ ✓
(2001)
CRISPR-Cas system (2012)
Reverse genetics for
✓ ✓ ✓
coronavirus (2003)
The case studies we conducted were:
1. Development of SARS animal model (“SARS-AM”), published in 2003,1041
2. Identification of high virulence mutation in H5N1 influenza virus (“Flu-PB2”), published in

2001,1042 and
3. Reconstruction of 1918 Spanish influenza strain (“Flu-1918”), published in 2005.1043
To ensure that the articles referenced above represented the starting point for proliferation, we examined
all similar hits that were generated by PubMed and WoS prior to the publication year. The search yielded
several hits and each was examined. We found that all SARS papers were on unrelated topics. In contrast,
several papers involving PB2 gene and 1918 influenza strain appeared relevant based on the abstract.
However, closer examination revealed that the PB2 papers implicated the region containing this gene as a
contributor to influenza pathogenesis, but did not pinpoint the gene itself.1044 The 1918 influenza papers
reported sequences of various RNA fragments, but not full
1041 Kuiken T et al (2003) Newly discovered coronavirus as the primary cause of severe acute respiratory syndrome. Lancet:
362(9380):263-70
1042 Hatta M et al (2001) Molecular basis for high virulence of Hong Kong H5N1 influenza A viruses. Science: 293(5536):1840-
2.
1043 Tumpey TM et al (2005) Characterization of the reconstructed 1918 Spanish influenza pandemic virus. Science:
10(5745):77-80.
1044 O'Neill E et al (2000) Heterologous protection against lethal A/HongKong/156/97 (H5N1) influenza virus infection in
C57BL/6 mice. J Gen Virol. 81(Pt 11):2689-96

reconstruction.1045,1046,1047,1048,1049,1050,1051,1052,1053,1054,1055,1056 Based on this analysis, we concluded that the
papers by Kuiken et al, Hatta et al, and Tumpey et al represented the first published reports for our case
studies.
10.5.5 Proliferation Trends
Extensive bibliometric searches were conducted to trace the propagation of the case study discoveries in
the research community. After combining and de-duplicating the publication sets we were left with 1,027
articles for SARS-AM, 685 for Flu-PB2, and 479 for Flu-1918. The titles and abstracts of these articles
were examined to determine whether they involved infection of animals with relevant strains or
manipulation of live virus, as well as containment levels of BSL-2+ or higher. After this process, the final
set contained 138 SARS-AM, 132 Flu-PB2, and 29 Flu-1918 papers, excluding the three initial discovery
articles. Note that at this stage we did not exclude groups working outside of the United States if they
published in English.
To characterize the level and rate of proliferation, we examined the number of papers published per year
and the number of groups (defined by the last author) publishing these papers. Each group was given one
count per year regardless of the number of papers they published in that year, but was counted in each
year they published a paper(s). For example, if group A published 10 papers in 2005, 5 papers in 2006,
and 1 paper in 2007, they will receive one count for 2005, 2006, and 2007.
As can be seen from Figure 10.2 (A-C), three discoveries followed different proliferation paths. The
SARS animal model was quickly “picked up” by about 20 groups, but the number of publications began
to decline in a few years. The chart shows that proliferation might be on the upward trend again, but not
enough data is available at present to determine whether this recent trend is anomalous. In contrast, the
uptake of the PB2 discovery has been slow, but continues to increase. Finally, the number of groups
working on the 1918 strain has remained low and constant.
1045 (a) Reid AH et al (1999) Origin and evolution of the 1918 "Spanish" influenza virus hemagglutinin gene. Proc Natl Acad Sci
U S A. 96(4):1651-6.
1046 Reid AH et al (2000) Characterization of the 1918 "Spanish" influenza virus matrix gene segment. Proc Natl Acad Sci U S A
97(12):6785-90.
1047 Reid AH et al (2002) Characterization of the 1918 "Spanish" influenza virus matrix gene segment. J Virol. 76(21):10717-
23.
1048 Reid AH et al (2003) Relationship of pre-1918 avian influenza HA and NP sequences to subsequent avian influenza strains.
Avian Dis. 47(3 Suppl):921-5.
1049 Reid AH et al (2004) Novel origin of the 1918 pandemic influenza virus nucleoprotein gene. J Virol. 78(22):12462-70.
1050 Basler CF et al (2001) Sequence of the 1918 pandemic influenza virus nonstructural gene (NS) segment and characterization
of recombinant viruses bearing the 1918 NS genes. Proc Natl Acad Sci U S A. 98(5):2746-51.
1051 Taubenberger JK et al (2001) Integrating historical, clinical and molecular genetic data in order to explain the origin and
virulence of the 1918 Spanish influenza virus. Philos Trans R Soc Lond B Biol Sci. 356(1416):1829-39.
1052 Gibbs MJ et al (2001) The haemagglutinin gene, but not the neuraminidase gene, of 'Spanish flu' was a recombinant. Philos
Trans R Soc Lond B Biol Sci. 356(1416):1845-55.
1053 Brownlee GG et al (2001) The predicted antigenicity of the haemagglutinin of the 1918 Spanish influenza pandemic
suggests an avian origin. Philos Trans R Soc Lond B Biol Sci. 356(1416):1871-6.
1054 Kobasa D et al (2004) Enhanced virulence of influenza A viruses with the haemagglutinin of the 1918 pandemic virus.
Nature. 431(7009):703-7.
1055 Tumpey TM et al (2004) Pathogenicity and immunogenicity of influenza viruses with genes from the 1918 pandemic virus.
Proc Natl Acad Sci U S A. 101(9):3166-71. Epub 2004 Feb 12.
1056 Reid AH et al (2003) 1918 influenza pandemic caused by highly conserved viruses with two receptor-binding variants.
Emerg Infect Dis. 9(10):1249-53.

A. SARS-AM B. Flu-PB2
Publications Groups Publications Groups
25
25
19 19
Number
Number
13 13
6 6
0 0
C. Flu-1918
Publications Groups
25
19
Number
13
Figure 10.2: Proliferation Trends for Each Case Study.
A number of factors might explain these trends, the most important of which is probably the levels of
federal funding, which are in turn dependent on perceived public health needs. In 2002, when the SARS
epidemic began, the NIH spent $4M on SARS research; this amount increased to $47M in 2004, to $82M
in 2008 and to $104 in 2010. Then the funding level began to decline and reached $84M in 2014. As can
be seen from Figure 10.2, the number of groups working on SARS also increased rapidly between 2004
and 2008. However, while funding for SARS remained high, the number of studies declined in 2009. The
reasons for this trend are unclear. It is possible that the researchers in this community encountered similar
experimental challenges, which have not yet been resolved. Since there were few of any new cases since
2004, the sense of public health need may have gradually declined. In 2012, the CDC added SARS to the
list of select agents, and while this occurred after the decline took place, increasing containment
requirements may be slowing research expansion in the past few years.1057 The increase in the number of
projects in 2014 might reflect the emergence of MERS, another coronavirus, and an associated increase in
research interest in SARS. However, at this time we do not have enough data points to determine whether
this uptick represents a change in trend.
In contrast to SARS, influenza outbreaks remain a widely discussed public threat, and this might be one
reason for the continued proliferation of the FLU-PB2 discovery. Federal funding for influenza has also
increased dramatically, from $47M in 2000 to $654M in 2014, offering new research opportunities.
Finally, the nature of the discovery may have contributed to the trend as well: it appears that many of the
1057 Federal Register Vol. 77, No. 194. Friday, October 5, 2012.

papers that followed the initial report tested the PB2 mutation in different influenza strains and animal
models.
Finally, we found no proliferation of the research reconstructing the 1918 influenza virus. This strain was
added to the list of select agents immediately after it was reconstructed in 2005, which probably inhibited
proliferation.1058
In addition to examining the proliferation trend, we calculated the total number of groups working in each
area, since the discovery was made through 2014. The total number of groups was 85 for SARS-AM, 64
for Flu-PB2 mutation, and 12 for Flu-1918 (Figure 10.3). Excluding all authors on the initial discovery
papers produces the estimates of research uptake by new groups, which was 70 for SARS-AM, 59 for
Flu-PB2, and 1 for Flu-1918. Finally, the number of groups was much lower for SARS and PB2 when all
of the papers whose last authors are based outside of the US were excluded. Mean funding levels over
2002-2014 were higher for the established than for new groups, at $29M versus $16M (p<.05, data not
shown).
All groups New groups, all New groups, US

85
70
Number of Groups
64
59
34
17
12
1 1
SARS-AM FLU-PB2 FLU-1918

Figure 10.3: Total Number of Groups Performing Experiments in the Case Studies
Table 10.6 shows last authors with at least six publications. Please note that all of the US-based
researchers in the table have also emerged as members of the interested GOF community (Table 10.3) and
as recipients of the NIH funding for influenza and SARS/MERS research (data not shown).
1058 Federal Register Vol. 70, No. 202 Thursday, October 20, 2005.

Table 10.6. Most Productive Authors in the Case Studies
Author Number of publications Case study
Y Kawaoka, University of
19 Flu-PB2, Flu -1918
Wisconsin
TM Tumpey, Centers for Disease
14 Flu -PB2, Flu -1918
Control and Prevention
RS Baric, University of North
12 SARS-AM
Carolina
K Subbarao, National Institute for
12 SARS-AM, Flu -PB2
Allergy and Infectious Diseases
HL Chen, Harbin Veterinary
9 Flu -PB2
Research Institute (China)
S Perlman, University of Iowa 9 SARS-AM
MG Katze, University of
7 Flu -1918
Washington
XF Liu, Yangzhou University
7 Flu -PB2
(China)
C Qin, Academy of Sciences
6 Flu -PB2
(China)
JK Taubenberger, National Institute
6 Flu -PB2, Flu -1918
for Allergy and Infectious Diseases
RG Webster, St Jude Hospital 6 Flu -PB2
Finally, we examined how the discoveries spread through the community. Were they taken up by new
researchers with no links to the authors on the initial papers? Or did they propagate via a narrow group of
the inventors’ students and collaborators? To answer these questions, we performed a network analysis, as
follows. If a last author on paper A became a middle author on a subsequent paper with the last author B,
then author A was considered to be the “parent” of author B. Our concept was that the investigator who
published earlier gave expertise, strains or laboratory space (in addition to experimental contributions) to
the laboratory that published later. For example, in Figure 10.4, the 2010 paper with Weingart as the last
author had Garcia-Sastre as middle author, so Garcia-Sastre is the parent of Weingart. By examining the
authorships using this principle, we constructed proliferation dendrograms for each case study (Figures
10.4-10.7).
Not surprisingly, we found that most authors were interconnected. For the smallest case study of Flu-
1918, a single author (Garcia-Sastre, the last author on the index paper) participated in the work of every
other last author (Figure 10.4). Similarly, for SARS-AM and Flu-PB2, three authors (Osterhaus/
Subbarao/Baric and Kawaoka/Tumpey/Webster, respectively) were key players in the propagation of their
research (Figures 10.5 and 10.6). The authors shown in bold and in light grey fonts are independent and
non-independent authors at the same institution as their parent, respectively.
Not all authors have left a lasting mark, however. Notice the dots on the bottom of panels B and C: these
are the last authors who published a single paper and left no “offspring,” at least at present. The fraction
of last authors that were connected to other last authors was 100% for Flu-1918, 44% for SARS-AM, and
59% for Flu-PB2. This analysis indicates that a good estimate of the proliferation potential in these fields

could be obtained by asking current researchers about their ongoing or planned collaborations and
determining if their senior post-docs plan to stay in the same field or move on.
Figure 10.4: Network Diagrams Showing Authorship Relationships for Flu-1918 Case Study.1059
1059 Each dot represents a paper with an indicated last author. If an earlier last author became a middle author on a subsequent
paper with a different last author, a line was drawn between the dots. PIs and non-PIs at the same institution are shown in
bold font and light gray font.

Figure 10.5: Network Diagrams Showing Authorship Relationships for SARS-AM Case Study.

Figure 10.6. Network Diagrams Showing Authorship Relationships for Flu-PB2 Case Study.
We repeated the analysis by examining whether middle authors became last authors on subsequent
papers. The concept here was that post-docs and graduate students in the lab of a last author would go on
to perform similar research and publish as senior author. While the specific author relationships appeared
different, the overall branching pattern held (see Appendix IV). These data suggest that a discovery
moves through the scientific community both through earlier groups giving rise to new groups (which
appeared to be the predominant pattern) and through the independent emergence of new groups.

10.5.6 Research Sites
We examined the sites in the United States where the GoF experiments published by the groups working
on SARS-AM, on Flu-PB2, and on Flu-1918 were performed. As most papers list several affiliations, we
made an assumption that the institution of the last author was the most likely experimental site. Figure
10.7 shows the number of institutions for each case study, across the studies and Table 10.7 lists the
facilities.
32
23
Number
13
7
SARS-AM FLU-PB2 FLU-1918 All

Figure 10.7: Number of Sites Performing Research Related to the Case Studies in the United States by
pathogen. An asterisk represents a known BSL-4 facility.
Table 10.7. Sites Performing Research Related to the Case Studies in the United States.
Battelle Memorial Institute State University of New York
Baylor University University of Alabama
Centers for Disease Control and Prevention University of Iowa
Diagnostic Center for Population and Animal Health University of Maryland
DynPort Vaccine Company University of Pittsburgh
East Carolina University University of Washington
Food and Drug Administration University of Wisconsin
Harvard University University of North Carolina
Johns Hopkins University University of California Berkeley
Kansas State University University of Rochester
Medical Research Institute for Infectious Diseases, Fort University of Texas Galveston*
Detrick*
Mount Sinai School of Medicine University of Central Florida
National Institute of Allergy and Infectious Diseases* US Department of Agriculture
Novavax Inc Utah State University
Stanford Research Institute Vanderbilt University
St Jude Children’s Hospital Yale University
*Represents a known BSL-4 facility

Finally, we reviewed the papers to identify the biosafety level of the facilities at which the experiments
were performed. Figure 10.8 shows that BSL-3 or BSL-3+ was the most common type in all three cases.
Note that 21 of 29 papers on 1918 influenza used BSL-3+ or BSL-4 facility, which as noted above can be
one explanation for the paucity of this research.
Unspecified BSL2+ BSL3 BSL3+ BSL4
2
5
29
Number of articles
97
71
1 3
34 31 18
7
1
SARS-AM FLU-PB2 FLU-1918
Figure 10.8: Number of Studies in each Case Study by BSL Level. Question mark indicates that the facility
level was not specified in the article and assumed to be BSL3.
10.5.7 NIH Funding
Please recall that we used “influenza, SARS, and MERS” terms and GoF PI names to query the NIH
database RePORTER in order to estimate the level of funding that might be available for GoF research.
As an independent measure of research support, we abstracted the grant numbers that were provided in
case study publications dated 2011-2015 and used these numbers to obtain funding amounts for that
period. Of the 61 grants and contracts referenced, the size of award was available for 32. The total
funding over the course of five years was $283M, or $57M per year. The median funding level for 14 PIs
was $783K, which was in the same range as what we found in previous analysis (Figure 10.7). Because
these estimates are based on the references within the GoF papers, we can conclude that some portion of
this funding was used on GoF research, but we cannot determine the specific amount.
10.6 Conclusions
We identified a group of 40 active, well-funded researchers in the US who have been performing, or have
the capacity to perform, certain types of GOF experiments involving influenza, MERS, and SARS
viruses. We also found that this research is currently not limited by access to containment facilities: in
fact, the existence of at least several hundred BSL-3 or higher level facilities offers potential for
expansion. This maximum number is supported by our case studies which showed that a new discovery in
this field may proliferate to as few as one and as many as 70 new groups around the world within 10-15
years. As indicated by the authorship patterns, about half of the new labs actively collaborate with
laboratories that published earlier whereas about half involve labs unconnected to these founders.
While establishing which of these paths a new discovery will take is impossible a priori, some
characteristics of the research seem reasonably correlated with greater or lesser proliferation potential. We
speculate that broader applicability of the discovery (for example a phenotype-conferring mutation in a

common gene or a new method) will facilitate proliferation and requirements for BSL-3+ or BSL-4
containment facilities will inhibit it, assuming that the number of these facilities does not increase.
Negative publicity associated with GoF experiments, additional regulatory oversight, public scrutiny of
the research, laboratory accidents, and uncertainty about future funding can limit proliferation.

11 Risk of Loss of Trust in Science
11.1 Summary
The majority of this document examines the risk to public health posed by the misuse of GoF research or
an accident at a facility conducting GoF research. However, after an incident of misuse of GoF research
or accidents involving a laboratory (e.g., loss of containment), loss of public trust is also a potential and
significant outcome. This loss of trust could arise via an accident that caused human illness or deaths, the
culling of livestock or wildlife, or even the perception of an increased hazard. Loss of public trust could
also arise via the publication of research that is perceived of having little benefit to the public but a great
potential for misuse, whether or not this perception is accurate. Overall, without direct polls indicating
reasons for loss of trust, assigning responsibility for loss of public trust to specific events is difficult.
The dynamic nature of trust is, however, open to analysis. By examining aspects of different types of
trust (e.g., contracts and regulations; the use of standards, certifications, or other assurance; the repetition
of positive events or interactions), case studies can be used to understand public loss of trust after specific
events originating from the scientific community. While we found no past incident that is a perfect
analogy for accidents that could occur during GoF experiments, the case studies on the Tuskegee Syphilis
Study and the Fukushima disaster demonstrate how people lost confidence in medical research among the
affected minority group or field (nuclear power). For the accidents at Bhopal and Pirbright, while no
quantitative sources were found, lawsuits and governmental action reflect a loss of public trust or increase
in concern. An increase in governmental regulation or oversight—as was seen in India after Bhopal—
reflects a loss of trust in the areas the regulation affects. Regulation and standards also reflect
governmental actions aimed at learning from or deriving some benefit from past events—like human
subject testing limitations after Tuskegee—in order to prevent them from happening in the future.
Employment and educational data were also examined to provide insight into how events shape the
choice of academic and career fields. However, these statistics may not paint an accurate picture due to
classification changes in both the academic programs and job categories. This limitation notwithstanding,
according to the employment and educational data, a correlation between the events studied and loss of
public trust in fields cannot be demonstrated.
From other scientific/ technological incidents examined, the outcomes included an increase in
government regulation, lawsuits, and, in the most devastating incidents, a long-term loss in public trust of
biomedical research. The longest-lasting harm observed arose from the Tuskegee Syphilis Study, which
still reduces African American participation in medical and research studies and affects models of health
care delivery to this population to the current day. The other examples presented may perhaps
demonstrate a loss of trust, but have seemingly had a minimal effect on research funding, foreign
investment, scientific education or scientific employment.
The concept of trust and its impact on behavior is a widely studied concept, but is generally defined as a
dynamic relationship “comprising the intention to accept vulnerability based upon positive expectations
of the intentions or behavior of another.”1060 Different forms of trust exist: deterrence-based trust, where
methods of control (e.g., contracts, regulations) come into play when sufficient trust is otherwise not
present; calculus-based trust, where the presence of control mechanisms is balanced with evidence of a
1060 Rousseau, D.M., Sitkin, S.B., Burt, R.S., and Camerer, C., Not So Different AfterAll: A Cross-Discipline View of Trust,
“Academy of Management, July 1, 1998, 23(3): 393-404, http://amr.aom.org/content/23/3/393.abstract

intentions or credibility (e.g., certifications, ‘trust-but-verify’); and relational trust, dependent on repeated,
positive interactions or the consistent meeting of expectations.1061 Thus, the public’s trust that the
biomedical research establishment can responsibly and safely conduct research on pathogens of pandemic
potential can be viewed as a dynamic relationship influenced by the contracts and regulations that are
established to control the research, assurances that the research is performed properly and are reinforced
by consistently meeting public’s expectations for the safety and security of the research. Additionally, in
terms of the public’s trust of science, loss of trust can be observed at three levels: the institution where the
accident occurred, the scientific field involved in the accident, and/or the scientific enterprise in general,
which could lead to long term consequences for research and development.
To provide decision-makers with some data on the risk of the loss of public trust in science due to
potential incidents involving GoF research, historic incidents were identified related to biomedicine,
science or technology and examined available information about the relevant determinants of trust (i.e.,
contracts and regulations, assurances, and meeting expectations) that reflected the public’s loss of trust in
scientific institutions, scientific fields, or the scientific enterprise in general.
Case studies were chosen on the basis of available data and applicable “lessons learned.”
Currently, most information on the topic of loss of public trust is not related to the creation of novel
strains of pathogens so it was necessary to analogize from other events. For this assessment, the
following incidents and/or accidents were considered:
• 1932-1972, the Tuskegee Syphilis Study in Alabama, USA,

• December 1984, Union Carbide Disaster in Bhopal, India,
• August 2007, Pirbright Foot-and-Mouth Disease Outbreak in Surrey, United Kingdom, and
• March 2011, Fukushima Daiichi nuclear disaster in Okuma, Fukushima, Japan.
Though there have been laboratory releases of agents –smallpox at the University of Birmingham
(19661062, 19781063), sabia at Yale University (19941064), tularemia at Boston University (20041065), SARS
at the National Institute of Virology in Beijing (20041066)—there is less information regarding public
opinion after these events which may perhaps be related to the small size or effect of these releases
compared to the events chosen for the case studies.
For this assessment, a variety of qualitative and quantitative sources regarding each of the above
described events were investigated:
• Primary and secondary media reports—including newspapers—provide a snapshot of public

reaction after an event, or on the anniversary of an event,
• Scholarly articles from academic journals provide reasoned feedback on the event as well as long-
term perspective on the impact of the event,
• Opinion polls measure public reaction quantitatively and provide data on how opinions change
over time, and
1061 Ibid.
1062 “Report of the Investigation into the Cause of the 1978 Birmingham Smallpox Occurrence,” (July 22, 1980): 30-34
1063 “Report of the Investigation into the Cause of the 1978 Birmingham Smallpox Occurrence,” July 22, 1980.
1064 “Scientist tests the public trust,” Nature 371 ( September 1, 1994): 1.
1065 Stephen Smith, “BU delayed reporting possibly lethal exposure,” Boston Globe (January 20, 2005)
1066 World Health Organization, “China Confirms SARS infection in another previous reported case; summary of cases to
date—Update 5,” (April 30, 2004) http://www.who.int/csr/don/2004_04_30/en/

• Lastly, congressional or governmental inquiries and the reports produced reflect public concern in
democratic countries.
Beyond the identified events, general studies concerning public trust and confidence in science were
gathered and analyzed. Statistics from the U.S. Department of Education’s National Center for Education
Statistics, Higher Education General Information Survey (HEGIS)1067,1068,1069 provide the number of
university science degrees conferred 1971-2013 and statistics from the National Science Foundation for
Science and Engineering Statistics’ NSF and NIH Survey of Graduate Students and Post-doctorates in
Science and Engineering provides the number of enrolled students for 1975-2011.1070 The data were
considered to identify interruptions of trends that correspond to the dates of historic accidents in order to
evaluate if accidents affect student completion of scientific degrees (BS, MS, and PhD). Employment data
comes from the U.S. Bureau of Labor and Statistics’ Occupational Outlook Handbook between 1972 and
2004.1071 Employment numbers can be considered estimates of the professionals working in a given field,
to determine if the historical events influenced overall employment in the field related to the incident.
11.3 Results
Overall, the data suggests that after an accident or disaster, the public is able to identify the responsible
party (for example, after the Fukushima Daiichi disaster, the most loss of public trust was suffered by
nuclear power) rather than blaming science generally—this is elaborated upon in each event specific
section, below. For each of the events outlined in this report, governmental action (e.g., governmental
inquiries or amendments to or the creation of legislation, lawsuits brought against the government, or
payment to affected individuals) reflected public concern but was set forth, largely, to prevent similar
events in the future. Although some incidents have led to measurable outcomes suggesting a loss of
public trust, an investigation of trends in enrollment and hiring in applicable STEM fields identified no
negative impacts following an event, which can be directly attributed to that event.
11.3.1 The Tuskegee Syphilis Study: Loss of Trust in Medical Research Among African Americans
The U.S. Public Health Service conducted an observational study of untreated syphilis in rural African
American men in Alabama from 1932-1972. The investigators did not fully disclose the nature of the
study to the participants, falsely told participants they were receiving treatment, and did not offer or
provide medical interventions/treatments when they became available. The victims of the study included
28 men who died as a direct result of syphilis, 100 men who died of complications related to syphilis, 40
wives of participants, and19 children born with congenital syphilis.
1067 U.S. Department of Education, National center for Education Statistics, “Table 322.10. Bachelor's degrees conferred by
postsecondary institutions, by field of study: Selected years, 1970-71 through 2011-12,”
https://nces.ed.gov/programs/digest/d13/tables/dt13_322.10.asp
1068 U.S. Department of Education, National Center for Education Statistics, “Table 323.10. Master's degrees conferred by
1069 U.S. Department of Education, National Center for Education Statistics, “Table 324.10. Doctor's degrees conferred by
1070 National Science Foundation “Survey of Graduate Students and Postdoctorates in Science and Engineering”
http://www.nsf.gov/statistics/srvygradpostdoc/
1071 U.S. Department of Labor, Bureau of Labor Statistics “Occupational Outlook Handbook,” Washington DC,
http://search.lib.virginia.edu/catalog/000046071

The public learned of the Tuskegee Syphilis Study after a 1972 article in the Washington Star exposed the
study. 1072 Prior to the 1972 article, two young physicians separately wrote to the PHS on three occasions
with ethical concerns regarding the study.1073,1074 One letter led the CDC to convene a blue ribbon panel
in 1969 that considered the study, recommending that the study be upgraded “scientifically,” yet decided
against treating the participants.1075 Of note, prior to this panel the U.S. Government led in the creation of
the Nuremberg Code (1942) to protect the rights of research subjects and generally recognized the
Declaration of Helsinki (1964), however this recognition of the ethical need for human subjects protection
was not reflected in the continued consideration or conduct of the Tuskegee Syphilis Study.
Since the 1972 article exposing the study, 1076 the U.S. government has attempted to make amends with
the public in a variety of ways. The NAACP filed a class action lawsuit on behalf of the study participants
seeking $3 million in damages for every living participant and the heirs of each participant. In December
1974, the U.S. Government settled out of court agreeing to pay $37,000 in damages to each survivor—
along with lifetime medical benefits for the survivors and any affected family members—and $15,000 for
the heirs of deceased study participants.1077 The aftermath of the Tuskegee Syphilis Study also led to
lasting changes in the conduct of research involving human subjects: in 1974 Congress passed and
enacted the National Research Act, which created the National Commission for the Protection of Human
Subjects of Biomedical and Behavioral Research. Despite monetary compensation and legislative
enactments, the U.S. government did not formally apologize for the Tuskegee Syphilis Study until 1997.
The legacy of the Tuskegee Syphilis Study is also thought by many to extend to the continued lack of
trust between the African American community and the U.S. medical system. In various studies of public
opinion since the closure of the Tuskegee study, African Americans point to these experiments as proof
that the medical research establishment and/or the U.S. government cannot be trusted in terms of equal
health care for all races or for providing informed consent; “The continuing legacy of the Tuskegee
Syphilis Study has contributed to Blacks’ belief that… public health authorities cannot be trusted.”
1078,1079,1080,1081,1082,1083
This distrust has been described by scholars dealing with the hesitance of African
1072 Abigail Perkiss, “Public Accountability and the Tuskegee Syphilis Experiments: A Restorative Justice Approach,” Journal
of African American Law and Policy 70 (2008): 71.
1073 Sarah Kaplan, “Dr. Irwin Schatz, the first, lonely voice against infamous Tuskegee study, dies at 83,” The Washington Post,
April 20, 2015, https://www.washingtonpost.com/news/morning-mix/wp/2015/04/20/dr-irwin-schatz-the-first-lonely-voice-
against-infamous-tuskegee-study-dies-at-83/
1074 Stephen B. Thomas, PhD, and Sandra Crouse Quinn, Med, “The Tuskegee Syphilis Study, 1932 to 1972: Implications for
HIV Education and AIDS Risk Education Programs in the Black Community,” American Journal of Public Health Vol 81,
No. 11 (1991): 1499, http://minority-health.pitt.edu/393/1/The_Tuskegee_Syphilis_Study_1932_to.pdf .
1075 Ibid.
1076 Abigail Perkiss, “Public Accountability and the Tuskegee Syphilis Experiments: A Restorative Justice Approach,” Journal
of African American Law and Policy 70 (2008).
1077 Ibid.
1078 Giselle Corbie-Smith, MD, Stephen B. Thomas, PhD, Mark V. Williams, MD, and Sandra Moody-Ayers, MD, “Attitudes
and beliefs of African Americans Toward Participation in Medical Research,” Journal of General Internal Medicine 14
(1999).
1079 Vicki S. Friemuth, Sandra Crouse Quinn, Stephen B. Thomas, Galen Cole, Erik Zook, and Ted Duncan, “African
Americans’ Views on Research and the Tuskegee Syphilis Study,” Social Science and Medicine 52 (2001): 797-808.
1080 Benjamin R. Bates, PhD, and Tina M. Harris, PhD, “The Tuskegee Study of Untreated Syphilis and Public Perceptions of
Biomedical Research: A Focus Group Study,” Journal of the National Medical Association Vol. 96, No. 8 (August 2004):
1051-1064.
1081 Bernard Lee Green, Richard Maisiak, Min Qi Wang, Marcia F. Britt, and Nonie Ebeling, “Participation in Health Education,
Health Promotion, and Health Research by African Americans: Effects of the Tuskegee Syphilis Experiment,” Journal of
Health Education 28 (1997)
1082 Vickie L. Shavers, PhD, Charles F. Lynch, and Leon F. Burmeister, “Knowledge of the Tuskegee Study and its Impact on
the Willingness to Participate in Medical Research Studies,” Journal of the National Medical Association 92 (2000)
1083 Stephen B. Thomas, PhD, and Sandra Crouse Quinn, Med, “The Tuskegee Syphilis Study, 1932 to 1972: Implications for
HIV Education and AIDS Risk Education Programs in the Black Community,” American Journal of Public Health Vol 81,
No. 11 (1991): 1499, http://minority-health.pitt.edu/393/1/The_Tuskegee_Syphilis_Study_1932_to.pdf .

Americans to participate in medical research generally and HIV/AIDS research specifically. 1084,1085,1086 In
her study of the Tuskegee Syphilis Study, Abigail Perkiss says, the “United States government had
committed gross injustices against members of the African-American community, and that community as
a whole was now beset by rampant distrust and suspicion toward the government and the medical
profession.”1087 This loss of trust in the biomedical research enterprise in general illustrates how loss of
trust from a particular incident may harm trust in biomedical research in general, despite institutionalized
changes to address the cause of the incident.
11.3.2 Bhopal Chemical Disaster
Arguably the worst industrial accident in history occurred in Bhopal, India where Union Carbide built and
operated a pesticide manufacturing plant. On December 3, 1984, more than 40 tons of methyl isocyanate
gas was released into the atmosphere, killing nearly 4,0001088 people instantly and harming the health of
an additional 15,0001089 to 600,0001090 people from acute and long term effects.
Investigations into the causes of the disaster found evidence of violations to operating procedures, as well
as a damning report from a 1982 safety inspection conducted by representatives from the U.S.-based
Union Carbide. 1091 This report indicated serious safety problems at the Bhopal plant and recommended
replacing one of the plants main safety devices (water spray system). In addition, research into the causes
of the disaster describe a state of confusion over ultimate responsibility for the plant’s and the public’s
safety, questions of legal accountability, and a poor safety culture and low morale among staff at the
plant.
Nearly immediately after the chemical release, those affected sought legal recourse both in American and
Indian courts of law. The cases brought in the U.S. were dismissed with the commentary that Indian
courts could better deal with these issues. However, neither Union Carbide, nor DOW Chemical (the
owner of what was formerly Union Carbide), have ever formally taken responsibility for the accident at
Bhopal, and have repeatedly placed the blame on the Indian staff at the plant. No one at DOW has been
held criminally liable, though in 2010 eight Indian mid-level managers were convicted of criminal
negligence. The government of India enacted the Bhopal Gas Leak Disaster Act as a way of “ensuring
that claims arising from the disaster would be dealt with speedily and equitably.”1092 The eventual legal
settlement in India was $470 million which would be paid to claimants as part of a full and final
settlement. In 2003, the Bhopal Gas Tragedy Relief and Rehabilitation Department reported that
monetary relief had been awarded to 554,895 people for injuries sustained in the disaster and to survivors
1084 Sengupta S, et. al. (2000) “Factors Affecting African-American Participation in AIDS Research,” Journal of Acquired
Immune Deficiency Syndromes 24: 275-284.
1085 Thomas S, Quinn S (1991) “The Tuskegee Syphilis Study, 1932 to 1972: Implications for HIV Education and AIDS Risk
Education Programs in the Black Community,” American Journal of Public Health Vol 81, No. 11
1086 Hagen K (2005) “Bad Blood: The Tuskegee Syphilis Study and Legacy Recruitment for Experimental AIDS Vaccines,”
New Directions for Adult and Continuing Education 105
1087 Perkiss A (2008) “Public Accountability and the Tuskegee Syphilis Experiments: A Restorative Justice Approach,” Journal
of African American Law and Policy 70: 72-73.
1088 Broughton E (2005) “The Bhopal disaster and its aftermath: a review,” Environmental Health: A Global Access Science
Source 4.
1089 Ibid.
1090 Malik A (2014) “30 Years After the Bhopal Disaster, India had not Learned the Lessons of the World’s Worst Industrial
Tragedy,” International Business Times.
1091 Diamond S (1985) The Phopal Disaster: How it Happened, The New York Times.
http://www.nytimes.com/1985/01/28/world/the-bhopal-disaster-how-it-happened.html?pagewanted=all .
1092 Broughton E (2005) “The Bhopal disaster and its aftermath: a review,” Environmental Health: A Global Access Science
Source 4.

of 15,310 who were killed. The average award amount was $2,200 for families of those killed1093 and
$400 for those who survived.1094
Beyond financial compensation, the governments of the United States and India passed legislation in
reaction to Bhopal. In 1990 the United States Congress passed the Clean Air Act Amendments (CAAA)
sections of this legislation require factories and other businesses to develop plans to prevent accidental
releases of highly toxic chemicals. The CAAA also established the Chemical Safety Board, an
independent agency, that investigates and reports on accidental releases of toxic chemicals from industrial
factories.1095 In India, multiple new pieces of legislation were passed in response to Bhopal including the
Environment Protection Act of 1986, amendments to the Indian Factories Act1096 and the Air Act in 1987,
Hazardous Water Management and Handling Rules in 1989, and the Public Liability Insurance Act of
1991.1097 Together, these pieces of legislation provide a framework similar to what exists in the US and
enable the Indian government to react to and prevent future accidents like the one at Bhopal as well as
setting forth best practices for handling hazardous waste or running industrial factories.
The influence of the Bhopal disaster on governmental decision making in terms of increased regulation of
chemical plants or foreign companies operating within India could be reflected in the levels of foreign
direct investment (FDI) before and after the Bhopal Chemical Disaster—if the Indian government enacted
legislation making India less attractive to foreign businesses after Bhopal, a decrease in the rate of growth
for overall FDI investment levels could be expected.1098 Overall, while no obvious dip was observed in
total FDI in the years following the Bhopal disaster (not shown), we note that prior to 1985 investment in
the chemical and pharmaceutical industries was an increasing share of the country’s total FDI (blue line in
Figure 11.1). In contrast, the data available for 1987 (after the Bhopal Disaster) marks the beginning of a
decline in the chemicals and pharmaceuticals sector’s share of total FDI relative to others. The rapid
growth of FDI in India’s chemicals and pharmaceuticals sector during this time period appears to have
settled a bit prior to the 1884 Bhopal incident, however the 1987 data indicate a distinct dip in the percent
increase in the sector’s FDI. That being said, although this sector captured a somewhat smaller share of
FDI after Bhopal, the sector still experienced a 100% annual growth rate after Bhopal (red line in Figure
11.1). Although these data represent merely a correlation with any incident, the reaction to the events at
Bhopal may have resulted in a decrease in foreign investment in India’s chemical and pharmaceutical
sectors.
1093 Ibid.
1094 Malik A (2014) “30 Years After the Bhopal Disaster, India had not Learned the Lessons of the World’s Worst Industrial
Tragedy,” International Business Times.
1095 United States Environmental Protection Agency, “The Plain English Guide to the Clean Air Act” (April 2007): 17.
1096 “The Factories Act, 1948 (Act No. 63 of 1948), as amended by the Factories (Amendment) Act 1987 (Act 20 of 1987)”
https://www.ilo.org/dyn/natlex/docs/WEBTEXT/32063/64873/E87IND01.htm
1097 Mannan M, et. al. (2005) “The legacy of Bhopal: The impact over the last 20 years and future direction,” Journal of Loss
Prevention in the Process Industries 18: 221.
1098 http://data.worldbank.org/indicator/BX.KLT.DINV.CD.WD/countries

40% 350%
35% 300%
30%
250%
25%
Bhopal Disaster 200%
20%
150%
15%
100%
10%
5% 50%
0% 0%
1948 1960 1970 1980 1987 1992 1995
Share of Chemicals & Pharmaceuticals, as % of total manufacturing FDI

Percent Increase in FDI in Chemicals & Pharmaceuticals Sector
Figure 11.1 Chemicals and pharmaceutical FDI as a percentage of overall manufacturing FDI and percent
increase in FDI in the chemicals and pharmaceuticals sector. 1099 Approximate timing of the Bhopal Chemical
Disaster is shown.
11.3.3 Pirbright FMD Outbreak
In August 2007 there were multiple outbreaks of Foot and Mouth Disease (FMD) among cattle herds in
Surrey, England. Well aware of the impact of FMD from the 2001 outbreak, farmers were forced to cull
their animals quickly to stem the spread of the disease. Overall, as the outbreaks began, the affected
public identified many possible environmental sources—migrating geese, local deer, or dogs.1100 Analysis
of the virus indicated that it was a strain of FMD isolated from the 1967 outbreak and used as a vaccine
strain by the nearby Pirbright Institute, which housed both the Institute for Animal Health (IAH, a
government diagnostic, research, and international reference lab run by the Department for Environment,
Food & Rural Affairs) and Merial Animal Health Ltd (a vaccine manufacturing factory). The consensus
of numerous governmental inquiries and commissions1101,1102,1103,1104 was that the pathogen was likely
accidentally released from the decades-old Pirbright facilities1105 through leaking effluent pipes and a
1099 Adapted from Suma Athreye and Sandeep Kapur, “Private Foreign Investment in India,” August 1999,
http://www.bbk.ac.uk/ems/faculty/kapur/personal/fdi.pdf
1100 Gray R (2007) “National Trust Estate Hit by Foot and Mouth,” The Telegraph.
1101 Return to an Address of the Honourable House of Commons, “Foot and Mouth Disease 2007: A Review and Lessons
Learned” March 11, 2008.
1102 Department of Environment, Food, and Rural Affairs (2007) “A Review of the Regulatory Framework for Handling Animal
Pathogens”.
1103 House of Commons, Innovation, University, Science, and Skills Committee, “Biosecurity in UK Research Laboratories,”
June 28. 2008.
1104 Health and Safety Executive, “Final Report on Potential Breaches of Biosecurity at the Pribright site 2007,” December 20,
2007.
1105 Department of Environment, Food, and Rural Affairs (2007) “A Review of the Regulatory Framework for Handling Animal
Pathogens,”: iii.

faulty valve at the Merial vaccine manufacturing plant.1106 That the pathogen was accidentally—as
opposed to intentionally—released may offer the closest example of scientific accident to one that could
occur while conducting GoF research in the United States.
Farmers who were affected by the depopulation of livestock brought a £1.5 million lawsuit against the
Institute for Animal Health and Merial Animal Health, as well as the Secretary of DEFRA.1107 IAH and
Merial settled with half of the farmers, while admitting no liability, and a judge dismissed the claims of
the other half (since none of their animals had been culled).1108
IAH at Pirbright is a critical facility for work with dangerous animal pathogens in the UK, and prior to the
2007 FMD release, the labs were due to be updated. The FMD accident did not prevent the renovation of
the research facility for the IAH which was approved in July 20091109 at a cost of £137 million. However,
as general budget discussions, and austerity measures, were implemented in the UK, the funding for
Pirbright, and three other priority government funded science projects, were cancelled.1110 Under these
austerity measures, funding was cut throughout the government—for health, business, local governments,
etc—not just for science.1111 Funding for the redevelopment of Pirbright was reorganized and covered
between the Biotechnology and Biological Sciences Research Council (BBSRC), DEFRA, and the
Department for Innovation Universities and Skills. At the time of this writing, the renovations are still
ongoing at Pirbright, however the new biocontainment facilities are already in use.1112 In short, no data
was found conclusively tying any negative consequences to science from this incident.
11.3.4 Fukushima Daiichi: Re-evaluation of Nuclear Power Worldwide
In March 2011, the massive Great East Japan Earthquake triggered a tsunami that disrupted cooling
systems at the Fukushima Daiichi Nuclear Power Plant resulting in several core meltdowns and damage
to the spent fuel. Radioactive material was released from the three affected reactors and people living
within a 30km radius of the plant were evacuated.1113 The financial cost of Japan’s recovery from the
Fukushima disaster is still ongoing, with nearly 250,000 Japanese still displaced1114 and the country
importing 90% of its energy.1115
Retrospective analyses of the factors that led to the Fukushima accident abound, including an influential
report from the Fukushima Nuclear Accident Independent Investigation Commission (NAIIC), an
independent body created by the National Diet (Japan’s parliament). The findings of the report point to
“a multitude of errors and willful negligence that left the Fukushima plant unprepared for the events of
March 11” as well as “serious deficiencies in the response to the accident by TEPCO, regulators and the
1106 Return to an Address of the Honourable House of Commons, “Food and Mouth Disease 2007: A Review and Lessons
Learned,” March 11, 2008: 12.
1107 Balakrishnan A (2008) “Farmers sue for damages in Pirbright foot-and-mouth outbreak,” The Guardian.
http://www.theguardian.com/uk/2008/oct/17/footandmouth-ruralaffairs
1108 “Foot-and-mouth cash demand fails,” BBC, March 31, 2009, http://news.bbc.co.uk/2/hi/uk_news/7974982.stm
1109 “Spending Review: Pirbright research lab escapes cuts,” BBC, October 20, 2010, http://www.bbc.com/news/uk-england-
surrey-11588361
1110 Sample I (2011) “Research cuts will force scientists to share laboratories, top academics warn,” The Guardian,
http://www.theguardian.com/science/2011/may/11/cuts-endanger-science-research-teaching
1111 “Spending Review 2010: Key points at-a-glance,” BBC, October 21, 2010, http://www.bbc.com/news/uk-politics-11569160
1112 http://www.research.pirbright.ac.uk/redevelopment/ “Phase two…is expected to be complete around 2016.”
1113 Siegrist M, Visschers V (2013) “Acceptance of Nuclear Power: The Fukushima Effect,” Energy Policy 59: 112.
1114 Spitzer K (2015) “250,000 japanese still displaced 4 years after quake, USA Today,
http://www.usatoday.com/story/news/world/2015/03/09/japan-tsunami-radiation-fourth-anniversary-fukushima/24254887/
1115 Fukushima’s impact on Japans economy three years on, BBC News, March 11, 2014, http://www.bbc.com/news/business-
26524084 .

government.”1116 More specifically, TEPCO and NISA were both cited for inadequately assessing the
earthquake and tsunami hazards faced by the plant. For example, TEPCO’s modeling did not adequately
incorporate the IAEA-promulgated best practice of including historic and pre-historic (i.e., evidence from
Japan’s geological record) seismic events and tsunamis. The committee also noted that a 2008 study by
TEPCO itself suggested that the tsunami hazard was greatly underestimated, however the company never
followed-up on this finding. As a result of the failure to take historical event into account meant that the
Fukushima plant was not designed to withstand a tsunami of even half the magnitude of the March 2011
event.1117 NAIIC Chairman Kiyoshi Kurokawa stated that “nuclear power became an unstoppable force”
in Japan which was “immune to scrutiny by civil society.” He continues that “Japan’s nuclear industry
managed to avoid absorbing the critical lessons learned from Three Mile Island and Chernobyl.”1118
The long-term effects of the Fukushima Daiichi nuclear disaster reached far beyond Japan’s borders. In
the days after the disaster, and with shaken confidence in nuclear power, governments around the world
performed tests and checks on their own reactors, took reactors offline, or started dialogues about the
future of nuclear power in their country. The European Union called for voluntary stress tests on reactors
within the EU and member countries reacted in various ways—Germany, with 17 reactors1119, shut down
the seven oldest, pending safety tests; Britain, with 19 reactors, and France, with 58 reactors, planned
safety reviews but decided not to delay nuclear expansion plans; Poland and the Czech Republic were
unaffected by the disaster and planned to continue with their nuclear plan development. Outside of
Europe, China temporarily suspended work on the approximately two dozen reactors under construction,
planned checks for operating reactors, and considered changes to their long-term nuclear power expansion
plans. Other earthquake prone countries, like India and Turkey, continued their nuclear development
plans unaffected by the Fukushima disaster.1120
Though caused by a “natural” event, the reaction of Japan’s public reflected a “radical alteration of …
a[n] optimistic view on science in policy making,”1121 a loss of public trust in both the impartiality of
scientists, of science in general.1122 Farther afield, the Pew Research Center,1123 conducted telephone
interviews in the United States immediately after the Fukushima accident to assess opinions on nuclear
power issues. In March 2011 39% of those polled favored promotion of increased nuclear power use
while 52% opposed. As a comparison, in October 2010, 45% favored and 44% opposed, demonstrating
that even though the accident did not occur in the US, 6-8% of Americans views of nuclear power became
more negative. Ipsos Global @dvisor, likewise, conducted a survey in May 2011 in 24 countries. Ipsos
found that countries in South and Southeast Asia—including South Korea, Japan, China, and India—
identified recent events in Japan as the source of their opposition to nuclear power.1124
Financial remuneration was offered to those living in Fukushima who were affected by the disaster. The
total bill for clean-up and remuneration of displaced residents is currently estimated at $137 billion
(USD), total, with costs to be covered by government issued bonds that TEPCO will repay over
1116 The National Diet of Japan “The Official Report of the Fukushima Nuclear Accident Independent Investigation
Commission,” 2012: 9.
1117 Carnegie endowment report.
1118 The National Diet of Japan “The Official Report of the Fukushima Nuclear Accident Independent Investigation
Commission,” 2012: 9.
1119 Kim Y, Kim M, Kim W (2013) “Effect of the Fukushima nuclear disaster on global public acceptance of nuclear energy,”
Energy Policy 61: 822-823.
1120 “Fukushima fall-out for reactors around the world” Nature.com, March 21, 2011.
http://blogs.nature.com.mutex.gmu.edu/news/2011/03/fukushima_fallout_for_reactors.html.
1121 Arimoto T, Sato Y (2012) “Rebuilding Public Trust in Science for Policy-Making,” Science 337: 1176.
1122 Ibid.
1123 Pew Research Center, “Opposition to Nuclear Power Rises Amid Japanese Crisis” March 21, 2011, 2-3.
1124 Ipsos—Global @dvisor, “Global Citizen reaction to the Fukushima Nuclear Plant Disaster” June 2011, 5.

time.1125,1126 While the event has caused a drag on the Japanese economy and affected public opinion on
the safety of nuclear technology worldwide, the lasting impact of Fukushima may be to highlight the need
to revise risk calculations, and the resulting safety margins, with current knowledge. As stated by James
Acton and Mark Hibbs for the Carnegie Endowment for International Peace, “In the final analysis, the
Fukushima accident does not reveal a previously unknown fatal flaw associated with nuclear power.
Rather, it underscores the importance of periodically reevaluating plant safety in light of dynamic external
threats and of evolving best practices, as well as the need for an effective regulator to oversee this
process.”1127
11.3.5 Effect of Incidents on U.S. Scientific Education
It was hypothesized that if any of these significant events harmed the U.S. public’s perception of science,
fewer students would be attracted to, enroll in, and later complete, degrees in related fields. The U.S. was
the focus of this study because the relevant data was available in English and the current study examines
the effect of U.S. action on risk of GoF research. It is recognized that a student in a degree program may
not drop the program due to a scientific accident, so the numbers may not drop immediately after an event
(if they drop at all). To control for economic factors that may influence the overall enrollment in post-
secondary education, the focus was on the percent of students that enroll or complete a degree compared
to all those enrolling in post-secondary education. National data showing the percentage of scientific
degrees earned (Figure 11.2 and 11.3) from the total number of degrees earned show no dips that could be
attributable to any particular incident/accident—physical sciences degrees have remained steady and
biological and biomedical degrees increased around the mid-2000s. The dip observed in Ph.D. completion
after the Tuskegee experiments occurs too soon after the revelation of the experiments to be attributable
to this event. Of note, an increase in the completion of undergraduate degrees in the life sciences is seen
four years after the 1972 revelation of the experiments. This uptick in life sciences undergraduate degrees
earned takes place two years after the influential 1975 Asilomar Conference on Recombinant DNA
which increased public interest in genetics and biomedical research.1128 Perhaps the decline in physical
science degrees since the post-Sputnik surge is partially attributable to the Bhopal disaster, but the
downward trend in physics Ph.Ds granted by U.S. institutions began before the incident.1129 Chemical
engineering has remained strong1130 and well represented in the engineering field (Figure 11.4), while
nuclear and biomedical remain steady with relatively low enrollment, non-respective of scientific
accidents.
1125 Inajima T, Song Y (2012) $137 Billion Cost has Tepco Seeking more Aid, Bloomberg Business.
http://www.bloomberg.com/news/articles/2012-11-07/fukushima-137-billion-cost-has-tepco-seeking-more-aid.
1126 Also, Catherine Butler, Karen A. Parkhill, and Nicholar F. Pidgeon, “Nuclear Power After Japan: The Social Dimensions,
Environment: Science and Policy for Sustainable Development 53:6 (2011): 6.
1127 Carnegie endowment report.
1128 Berg P, Singer M (1995) “The recombinant DNA controversy: Twenty years later”, PNAS,
http://www.pnas.org/content/92/20/9011.
1129 Kaiser D ( American Physics and the Cold War Bubble, (University of Chicago Press, in preparation),
http://web.mit.edu/dikaiser/www/CWB.html .
1130 It is unclear what the large drop between 2002 and 2003 can be attributed to. Based on the way these studies are conducted,
it is likely attributed to a reclassification of the degree or program type, however, no such information explaining this was
found within the National Science Foundation’s records.

7.00%
6.00%
Percentage of Overall Degrees Earned
5.00%
4.00% Biological and

Fukushima Biomedical (BS)
3.00% Tuskegee Biological and

Syphilis Study Bhopal Biomedical (MS)
Chemical Pirbright
Disaster Biological and
2.00%
Biomedical
(PhD)
1.00%
0.00%
Academic Years, 1970-2013

Figure 11.2. Percent of degrees in the life sciences awarded by U.S. Institutions as a percent of the total degrees awarded. The timing of the catastrophic
events is shown. Statistics drawn from the U.S. Department of Education’s National Center for Education Statistics, Higher Education General
Information Survey (HEGIS).1131,1132,1133
1131 U.S. Department of Education, National center for Education Statistics, “Table 322.10. Bachelor's degrees conferred by postsecondary institutions, by field of study: Selected
years, 1970-71 through 2011-12,” https://nces.ed.gov/programs/digest/d13/tables/dt13_322.10.asp
1132 U.S. Department of Education, National Center for Education Statistics, “Table 323.10. Master's degrees conferred by postsecondary institutions, by field of study: Selected
1133 U.S. Department of Education, National Center for Education Statistics, “Table 324.10. Doctor's degrees conferred by postsecondary institutions, by field of study: Selected

7.00%
6.00%
Percentage of Overall Degrees Earned
Fukushima
5.00%
Bhopal
Tuskegee Chemical
Syphilis
4.00%
Physical Science and
Science Technologies
(BS)
3.00% Physical Science and
Pirbright Science Technologies
(MS)
Physical Science and
2.00% Science Technologies
(PhD)
1.00%
0.00%
1988-89
1970-71
1972-73
1974-75
1976-77
1978-79
1980-81
1982-83
1984-85
1986-87
1990-91
1992-93
1994-95
1996-97
1998-99
2000-01
2002-03
2004-05
2006-07
2008-09
2010-11
2012-13
Academic Years, 1970-2013
Figure 11.3. Percent of degrees in the physical sciences awarded by U.S. institutions as a percent of the total degrees awarded. The timing of the
catastrophic events is shown. Statistics from the U.S. Department of Education’s National Center for Education Statistics, Higher Education General
Information Survey (HEGIS).1134,1135,1136
1134 U.S. Department of Education, National center for Education Statistics, “Table 322.10. Bachelor's degrees conferred by postsecondary institutions, by field of study: Selected
1135 U.S. Department of Education, National Center for Education Statistics, “Table 323.10. Master's degrees conferred by postsecondary institutions, by field of study: Selected
1136 U.S. Department of Education, National Center for Education Statistics, “Table 324.10. Doctor's degrees conferred by postsecondary institutions, by field of study: Selected

Figure 11.4. Percent of enrollments in graduate study in relevant fields at U.S. institutions as a percent of the total graduate enrollments. The timing of
the catastrophic events is shown. Statistics from the National Science Foundation for Science and Engineering Statistics’ NSF and NIH Survey of
Graduate Students and Post-doctorates in Science and Engineering. 1137
1137 National Science Foundation “Survey of Graduate Students and Postdoctorates in Science and Engineering” http://www.nsf.gov/statistics/srvygradpostdoc/

11.3.6 Effect of Incidents on Scientific Employment
Similar to the rationale described above, if any of the described events harmed public perception of
science, that data may show a drop in number of employees in relevant fields. In the case of employment
data, it is possible that an event would cause an immediate drop in employment if workers left their jobs
in disgust. The number employed in a variety of related fields (Figure 11.5) does not appear to be affected
by any event described in this report, though it may reflect economic factors not related to any particular
scientific accident or event. Moreover, any drop in employment could have been compensated for by the
filling of vacancies with employees on a work visa. Another possible explanation for drops in fields could
be shifting categories of employment (for example, Biological and Biomedical fields are combined in
some years and separate for others). In 1998, shifts upwards and downwards are likely a result of these
changing category designations.

90.00%
80.00%
70.00%
Bhopal
Chemical
Percentage of Total Field Employed as...
60.00%
Disaster
Biological
Scientists
50.00%
Chemists
Chemical
40.00% Engineers
Nuclear
Engineers
30.00%
Physicists
20.00%
10.00%
0.00%
1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2004
Year
Figure 11.5. Percent employed in relevant fields in the U.S. as a percent of engineers and scientists employed. Timing of applicable catastrophic events is
shown.

Nuclear and chemical engineers represent a small proportion of total employment numbers, but the
proportion remains steady throughout the time surveyed. We note that the nuclear accidents we examined
lie outside the timeframe of the data collected. Similarly, the Tuskegee Syphilis Study falls outside the
employment numbers, it is conceivable that the low number in 1978 could be attributed to the
experiments—however, it also could be attributed to the shifting categories reflecting employment. The
FMD outbreak at Pirbright also falls outside the years this data was collected.
11.3.7 General Opinion
One final method used to measure public trust in science were long-term and recently conducted general
opinion surveys about science conducted by the General Social Survey (GSS) and the Pew Research
Center. Figure 11.7 displays results from the General Social Survey which indicates slight increases or
decreases in U.S. public confidence in the scientific community attributable to no specific event. Both the
GSS and Pew results demonstrate that trust in science has remained consistent over the past 40 years.
Recent results from Pew indicate that general trust in science has decreased slightly from 20091138 to
20151139 but remains relatively high. These numbers show little if any effect from the Fukushima disaster
on the public’s opinion of science, in general, within the US. Pew, additionally, asks questions about
occupational fields, and scientists are seen as contributing “a lot” to society’s well being (only members
of the military and teachers are ranked higher than scientists; doctors and engineers rank similarly to
scientists.)
60.00%
Confidence" in the Scientific Community
Percent who answered "Great Deal of
50.00%
40.00%
Fukushima
Tuskegee Bhopal Pirbright Daiichi
30.00%
Syphillis Chemical FMD
Study Disaster
20.00%
10.00%
0.00%
1978
1997
2007
1973
1974
1975
1976
1977
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1998
1999
2000
2001
2002
2003
2004
2005
2006
2008
2009
2010
2011
2012
2013
2014
Year Surveyed
Figure 11.6. Percent of U.S. public surveyed answering they have a “great deal of confidence” in the scientific
community. The timing of the catastrophic events is shown. Data from the General Social Survey. 1140
1138 Pew Research Center, “Scientific Achievements Less Prominent Than a Decade Ago,” July 9, 2009.
1139 Pew Research Center, “Public and Scientists’ Views on Science and Society,” January 29, 2015.
1140 General Social Survey, NORC at the University of Chicago, http://www.norc.org/Research/Projects/Pages/general-social-
survey.aspx

12 Appendix I: Glossary
Absolute risk: risk given in terms of consequences per unit of real time
Adjuvants (for vaccines): substances that are added to a vaccine to boost or otherwise modify the
recipient’s immune response to the vaccine antigens, in order to enhance the vaccine’s effects
Antibody escape mutant: a virus that has acquired mutations at antigenic sites that prevents antibody
neutralization
Antigen: a substance that leads an immune system to generate antibodies against it
Antigenic drift: small changes in the antigenic character of a virus that alter antibody neutralization
Antigenic shift: the phenomenon of HA or NA gene segments being exchanged between viruses in
nature
Antiviral: compounds used to treat or prevent viral infections. Antivirals are a type of medical
countermeasure
Assay: a general term used to describe a range of laboratory techniques that determine or measure the
presence, amount, or activity of a particular substance
Atmospheric dispersion model: a model that predicts the transport of a contaminant in the air from a
release site
Attenuated: a pathogen that is still able to grow in its host, but has reduced virulence. Live attenuated
vaccines use an attenuated pathogen strain with extremely low virulence
Backbone (strain): an often-attenuated virus strain that contains internal gene segments of an influenza
virus. A/Puerto Rico/8/1934, A/WSN/1933, and A/Ann Arbor/6/1960 are commonly used backbone
strains
Biosafety: the concept of reducing the risk of natural or accidental exposure to pathogens. Biosafety
measures at a laboratory reduce the probability of accidental human exposure to an agent, and where
possible, reduce the consequences of such an event should it occur
Biosecurity: the concept of reducing the risk of deliberate exposure to pathogens. Biosecurity measures
at a laboratory seek to guard stored pathogens against theft, diversion, or other intentional misuse, and to
mitigate the consequences of such acts should they occur
Biosurveillance (surveillance): the systematic process of gathering and analyzing data that might relate
to pathogen activity in the hopes of detecting disease outbreaks
Branching process model: a model of a population in which each individual in a generation produces a
random number of offspring. In this report, branching process models are used to predict how many
infected individuals (offspring) are produced by any infected individual in a nascent outbreak.
"Bright line" boundary: an easily-applied objective rule that resolves an issue in a clear-cut manner

Biological Select Agent or Toxin: disease agents subject to federal oversight and regulation through the
Federal Select Agent Program. HHS pathogens and toxins are those deemed to pose a high risk to human
health, while the USDA pathogens and toxins are deemed to pose a high risk to plant or animal health.
Certain “overlap” pathogens are on both lists
Biosafety Level: a laboratory ranking system, as defined in the BMBL, that assigns a level to sets of
facility standards, laboratory practices, and types of safety equipment in terms of the overall containment
capacity provided. Biosafety levels range from Biosafety level 1 (lowest level of containment) to
Biosafety level 4 (highest level of containment)
Case Fatality Rate: the rate of deaths within the population of people infected with a pathogen
Cell culture: growing and maintaining cells isolated from an organism under controlled laboratory
conditions
Cell line: a population of cells descended from a single cell, generated and maintained through cell
culture
Chimeric microorganism: a microorganism created by joining nucleic acid fragments from two or more
microorganisms
Codon: a triplet of adjacent DNA or RNA nucleotides that together define a specific amino acid or a stop
signal during protein synthesis
Cognate antibody: an antibody corresponding to an antigenic site on the influenza virus, often used to
map the different viral epitopes or to generate adaptive immune response escape mutants
Convalescent sera: the sera isolated from an animal or human after infection that contains antibodies
specific to the infecting pathogen
Coronaviruses: common viruses belonging to two subfamilies (Coronavirinae and Torovirinae) of the
virus family Coronaviridae. In this report, the term is used to describe the SARS- and MERS-CoVs not
the coronaviruses that may cause the common cold.
CRISPR-Cas9: a recently developed laboratory technique used to modify genomic DNA
Crosswalking: mapping correspondences between two or more sets of data, for instance the mapping of
information to knowledge gaps
Dendrograms: tree diagrams often used to organize and display association between genes or samples by
grouping them into clusters
Dose-response models: models that predict the effect on an organism caused by increasing doses of a
stressor. In this report, dose-response models are used to predict the probability of infection caused by a
dose of a pathogen
Dose-sparing: an approach that seeks to reduce the amount of antigen required for effective vaccination
Downstream: events that occur further along in a process chain

Dual use research: a generic term that refers to civilian research that could be diverted to serve a military
purpose
Dual Use Research of Concern: the official term used by the U.S. government in documents on ensuring
institutional oversight of dual-use research in the life sciences, i.e. to ensure oversight of “life science
research that, based on current understanding, can be reasonably anticipated to provide knowledge,
information, products, or technologies that could be directly misapplied to pose a significant threat with
broad potential consequences to public health and safety, agricultural crops and other plants, animals, the
environment, materiel, or national security.”1141
Epitope: the part of an antigen to which an antibody binds
Fault tree analysis: an analytical technique in which pathways within a system that can lead to a
predictable failure are described using Boolean logic
FDA-qualified animal model: an animal that has been approved for laboratory use through the FDA’s
Animal Model Qualification Process to accurately represent human disease processes
Fitness: the ability for an organism to survive and reproduce
Fomite: a surface or object contaminated with pathogens that can therefore serve as a vehicle in disease
transmission
Forward genetic screen: the process of modifying genetic code, often a priori, to elucidate gene
sequences responsible for particular phenotypes
Gain of Function—Research/Experiments: laboratory experiments that are reasonably expected to

generate influenza or coronaviruses with enhanced growth, pathogenicity, transmission between
mammals, vaccine evasion, or resistance to medical countermeasures
Gain of Function—Pathogens: the organisms subjected to gain of function experiments. In this report,
these organisms are influenza A viruses, and SARS and MERS coronaviruses
Gain of Function—Laboratory: a workplace where gain of function experiments take place
(Genetic) modifications: an alteration of genetic material. See Mutation
Genotype: the genetic makeup of an organism
Hemagglutinin (HA): the viral protein from influenza virus that causes red blood cells to agglutinate
Hemagglutination inhibition (HI, HAI) assay: evaluates the ability for antibodies to prevent a virus
from agglutinating red blood cells
Host immune modulators: substances that alter typical immune function
Host tropism: a pathogen with improved fitness in a specific host relative to other hosts
1141
Sorrell EM et al (2009) Minimal molecular constraints for respiratory droplet transmission of an avian-human
H9N2 influenza A virus. Proc Natl Acad Sci U S A 106: 7565-7570

in ovo: in chicken eggs
in vitro: in cell culture or in cell-free biochemical systems
in vivo: in humans or other animals
Inactivated: refers to a pathogen that has been rendered non-infectious
Influenza: a disease caused by the influenza virus
Isolate: a pure strain of a pathogen separated from a mixed culture
K: a variable used in epidemiology, defined as the variation of infectiousness between individuals
Knockout cell lines: a cell line in which one or more proteins are not expressed due to a removal or
modification of the DNA encoding that protein
Markov chain model: a model in which the next state of a subject in the model is determined exclusively
by its current state (and not its prior history)
Medical countermeasures: vaccines, medications, or equipment used to improve public health outcomes
in response to a disease outbreak
Middle East Respiratory Syndrome: a respiratory illness caused by a coronavirus, MERS-CoV. The
first known cases of MERS were reported in Saudi Arabia in 2012
Monoclonal antibody: an antibody derived from a single B-cell line, which generates an antibody to one
epitope on an antigen
Monte Carlo simulation: a simulation in which the probability of various outcomes is predicted via the
analysis of multiple model runs, each of which use parameter values selected at random
Morbidity: the number of individuals exhibiting disease symptoms in a given population
Mortality: the number of deaths for a given population
Mutagenesis (mutagenizing): the process of changing one or more nucleotides of DNA
Mutation(s): a change of one or more DNA nucleotides
Myalgia: muscle pain
Neuraminidase inhibitors: chemical compounds that block the viral neuraminidase enzyme, preventing
virus replication. Neuraminidase inhibitors are currently used against influenza; examples of such
compounds mentioned in the report include zaninamivir, oseltamivir, peramivir, and laninamivir
Novel strain: a strain of a microbe distinct from any previously characterized strain
Orthomyxoviruses: a family of RNA viruses from six genera: influenza virus A, influenza virus B,
Influenza virus C, Isavirus, Thogotovirus, and Quaranjavirus.

Pandemic: an epidemic occurring worldwide or over a very large area and affecting a large number of
individuals
Parametric approach (parametric analysis): a modeling approach that uses multiple different
parameters, each with an accompanying finite range of potential values, to describe a range of
characteristics of a subject and their effect on model outcomes. In this report, a parametric approach was
used to describe a variety of pathogens with a range of phenotypes manipulated under undetermined
laboratory conditions to explore their influence on risk.
Parental strain: the original virus strain used as the basis for subsequent genetic modification, creating
novel strains
Passaging: the process of placing a virus strain under selective pressure in cells or animals for several
iterations to introduce adaptive mutations
Pathogenicity (pathogenesis): a characteristic of a virus or organism that generates harmful biological

responses in the host
Phenotypic (phenotype): the physiological or measurable result of a genotype; a trait
Polyclonal antibodies: a collection of antibodies taken from the serum of an immunized individual, each
of which may bind to a distinct epitope with a variety of strengths
Probabilistic risk assessment: a systematic method to assess risks, in terms of consequence and
probability, for complex systems
Prophylactically (prophylactic): medications taken to prevent infection and disease
R0: R0 is a variable used in epidemiology, defined as the reproductive number of a transmissible pathogen
(see Transmissibility), or the number of infected cases one infected person will create
Random mutagenesis: a laboratory technique that randomly generates mutations in DNA
Reagent: a substance or mixture of substances used in an assay or other laboratory technique
Reassortment (reassortant) (reassorting): a laboratory technique used to exchange gene segments

between two or more viruses to generate new viruses. A 6:2 reassortment strain has six gene segments
came from one parental strain, and two gene segments came from the other parental strain to form the
new virus
Recombinant: recombining of genetic material, often from different sources, to generate new genetic
sequences
Relative risk: risk of a novel event compared to the risk of a baseline event. In this report, relative risk of
research on GoF pathogens is compared to risk of research on wild type pathogens (as the baseline).
Relative risk is provided when the frequency of a negative event is unpredictable.
Reservoir: any organism that typically harbors a pathogen. The pathogen depends on and grows in the
reservoir, and can subsequently infect other organisms in contact with the reservoir
Reverse Genetics: in this report, an approach that generates a virus from isolated genetic material

Risk: this report uses the actuarial definition of risk, the product of consequences arising from a negative
event and the probability of the negative event
SEIR model: an epidemiological model that tracks the flows of hosts from the susceptible state (S) to the
exposed state (E) to the infected (I) and resistant (R) states.
Serotype: viral strains described by the category of antigens displayed on the outside of the virus. Three
serotypes for influenza exist, including influenza A, B, and C
Severe Acute Respiratory Syndrome: a respiratory disease caused by the coronavirus SARS-CoV. The
first cases of the disease were reported in Asia in February 2003
Sialic acid moieties: in this report, residues expressed on the outside of cells which the viral HA
recognizes and binds to, allowing for infection
Site-directed mutagenesis: a laboratory technique that introduces specific mutations into DNA
Stochastic: characterized by a random probability that is statistically analyzable but not predictable a
priori
Sublineages: a group of related viruses descending from a common ancestor strain
Subtype: viral strains described by the set of HA and NA proteins expressed on the virus surface
Tier 1 BSAT: A select agent deemed to pose the greatest threat, which are subject to additional
regulatory safety and security requirements
Titer: a measure of a virus’ concentration in a given sample
Translators: the individuals who apply (“translate”) fundamental research results to practice (in this
report, basic research results to benefits in public health or medicine)
Transmissibility: the ability of a pathogen to spread from an initial case through a population
Vaccine (protective vaccination): a type of medical countermeasure that stimulates an immune response
to prevent or mitigate future infection
Vaccine platform (platform): a set of processes and methods used to generate vaccines
Virulence: the characteristic of a virus that informs the severity of morbidity and mortality
Vivarium: a part of a laboratory complex dedicated to housing research animals
Wild type virus (strains): an unmodified organism
Yield: the remaining amount of a substance after one or more processes
Zoonotic (disease): a disease that can infect lower animals and humans

13 Appendix II. Acronyms Used
ABSL animal biosafety level

ALF Animal Liberation Front
alt GoF alternatives to gain of function
APHIS Animal and Plant Health Investigation Service
BARDA US Biomedical Advanced Research and Development Authority
BMBL Biosafety in Microbiological and Biomedical Laboratories
BPM branching process model
BSAT biological select agents and toxins
BSL biosafety level
BW biological warfare
BWC Biological Weapons Convention
CBRN Chemical Biological Radiological Nuclear
CDC US Centers for Disease Control and Prevention
CFR case fatality rate
CFR Code of Federal Regulations
CoV coronavirus
CVV candidate vaccine virus
DALY disability-adjusted life years
DEFRA UK Department for Environment, Food, and Rural Affairs
DOD US Department of Defense
DURC dual use research of concern
EAR Export Administration Requirements/ Regulations
ELF Earth Liberation Front
EPI Emerging Infections Program
FBI US Federal Bureau of Investigation
FDA US Food and Drug Administration
FDI Foreign Direct Investment
FEMA US Federal Emergency Management Agency
FMD foot and mouth disease
FSAP Federal Select Agent Program
FTA fault tree analysis
GAO US Government Accountability Office
GDP gross domestic product
GoF gain of function
HA hemagglutinin
HEPA High-efficiency particulate air (filter)
HHS US Department of Health and Human Services
HPAC Hazard Prediction and Analysis Capability
HPAI highly pathogenic avian influenza
HVAC heating, ventilation, and air conditioning

IAH Institute of Animal Health (UK)
IBC Institutional Biosafety Committee
ID infectious dose
IIM interactive influenza model
IND Investigational New Drug
ISIL Islamic State of Iraq and the Levant
ITAR International Traffic in Arms Regulations
LAI lab acquired infection
LAV live attenuated viruses
LD lethal dose
LoF loss of function
LPAI low pathogenic avian influenza
MCM medical countermeasures
MERS Middle East Respiratory Syndrome
NA neuramidase
NAIIC Fukushima Nuclear Accident Independent Investigation Commission
NBAF National Bio- and Agro-Defense Facility
NEIDL National Emerging Infectious Diseases Laboratory
NIAID US National Institute of Allergy and Infectious Diseases
NIH US National Institutes of Health
NRC National Research Council
NSABB US National Science Advisory Board for Biosecurity
NSDD National Security Decision Directive
OSHA US Occupational Safety and Health Administration
P&I pneumonia and influenza
pfu plaque forming unit
PI principal investigator
PPE personal protective equipment
PPP pathogens with pandemic potential
PRA probabilistic risk assessment
RA risk assessment
RAC Recombinant DNA Advisory Committee
RBA risk benefit analysis
SAR select agent regulations
SARS Severe Acute Respiratory Syndrome
SARS-AM SARS animal model
SEIR susceptible, exposed, infectious, recovered
SME subject matter expert
SRA security risk assessments
START National Consortium for the Study of Terrorism and Responses to Terrorism
TEPCO Tokyo Electric Power Company
USDA US Department of Agriculture
USG US Government

VSVG vesicular stomatitis virus glycoprotein
WHO World Health Organization
WMD weapons of mass destruction
WoS Web of Science database

14 Appendix III. Additional Data on the Methods of the
Quantitative Risk Assessment
Additional Methodological Information Supporting the Estimate of Loss of
Containment Pathways 490
14.1.1 Elimination of implausible incidents leading to loss of containment 490
14.1.2 Elimination of Some Incident Pathways from Fault Tree Modeling 491
14.1.3 The Monte Carlo framework 493
14.1.4 Human Reliability Assessment in Biological Laboratories 495
14.1.5 Further Information on Modeling Infection Risk Caused by Fomites 501
14.1.6 Sources used to identify incident scenarios to include in the study 505
Methodological Details of the Branching Process Model 506

14.2.1 Probability Distribution Used in the Model 506
14.2.2 Construction of a Two-Type Model for Workers and Community Members 507
14.2.3 Incorporation of Control Measures 509
14.2.4 Terminating Models Due to Loss of Control 511
14.2.5 Calculation of Self-Extinguishing Probability 512
Methodological Details of the HHS-BARDA IIM 512

14.3.1 Computation of region-specific contact rate matrices 512
Additional Data on the Potential Proliferation of GoF Research 514

14.1 Additional Methodological Information Supporting the Estimate of Loss of
Containment Pathways
14.1.1 Elimination of implausible incidents leading to loss of containment
Once the list of incidents to investigate was finalized, pathways by which these incidents would lead to a
loss of containment were researched. In so doing, no plausible way was found for some incidents to lead
to a loss of containment and so these incidents were eliminated from quantitative modeling. These
implausible incidents are listed here.
14.1.1.1 Loss of Power should not occur in a containment laboratory
Requirements for high-containment laboratories stipulate that power must be supplied by two completely
independent conduits from two sources, suggesting that two, simultaneous power outages must occur.
Moreover, backup generator power is required. For this reason, a power outage would have to occur via
three, extremely unlikely events. Even if a power outage is experienced, louvers in place are designed to
fail safe and isolate the laboratory from the outside. Standard protocols require workers to immediately
cease and secure work (e.g. by closing the sashes on any active BSCs). This event requires four
completely independent, rare events to happen and therefore would be vanishingly unlikely. Also,
laboratory work would not continue in a power outage suggesting that there is very little opportunity for
an accident to occur during this period. Continuous sources of aerosols (animals in containment) are very
dilute and pose a minimal risk even if the power and the louvers fail (see animal aerosol risk, below).
14.1.1.2 Floods should not lead to a loss of containment
Some of the laboratories identified in our study are in areas of some flood risk (protected by levees or
not). Floods are not unanticipated events, and days of warning precede a flood caused by a tidal surge
from a hurricane or a river flood from excessive rain (none of the laboratories identified were in a gulley
susceptible to flash floods). As learned in the interviews, in anticipation of previous hurricanes (such as
hurricane Sandy) or other flooding events, the researchers sacrificed all infected animals, decontaminated
the laboratory and shut it down. Even if these measures were not taken, the risk of a loss of containment
from floods would be minimal. Firstly, the containment facilities of these laboratories are on the upper
floors of the building so would not actually be inundated by the flood. Power, which, by requirement,
must be supplied by two independent conduits and sources, would likely not be interrupted. For these
reasons, even in the rare instance of a significant flood striking a containment laboratory, practices and
laboratory configurations would eliminate the risk of a loss of containment.
14.1.1.3 Shipping accidents should not lead to a loss of containment with GoF pathogens
In our interviews with GoF laboratories, we found that samples of GoF pathogens are not shipped out of
the laboratory. Reverse genetics techniques are so routine in these laboratories that strains are “shared”
between labs by the sharing of high-fidelity sequence information, the synthesis of the viral genomes and
the rescue of the active viruses. Shipping is routinely used, however, in laboratories that analyze wild type
samples for these samples.
14.1.1.4 Improper inactivation should not lead to a loss of containment event with GoF pathogens
Because pathogens are not shipped from GoF laboratories, the only materials that are inactivated that are
taken out of the laboratory (not in the waste streams) are samples for analysis by molecular methods or
microscopy. These samples are decontaminated and fixed, and the samples are placed in boxes and
dunked into a decontaminant bath. There is no physical contact with the sample and the sample does not

leave the laboratory except through the waste stream. In addition, the inactivation procedures used here
typically destroy the virus entirely by, for example, formalin treatment for cell fixing or Trizol treatment
for RNA extraction. These procedures stand in contrast to inactivation procedures for other pathogens that
inactivate them but leave them intact, such as the radiation inactivation protocol for Bacillus anthracis. If,
however, some infectious material somehow ends up on a researchers glove during the procedure (or in
the waste stream) these events are captured in the splash or waste stream incidents.
14.1.2 Elimination of Some Incident Pathways from Fault Tree Modeling
After investigating some incidents for quantitative modeling, the events were found to be so infrequent or
so inconsequential (or both) that there was no need to include them in Fault Tree modeling because it was
predictable that these events would not contribute to the risk of a loss of containment accident. The
process for excluding those events is described here.
14.1.2.1 Liquid Waste Disposal
In this scenario, untreated liquid waste containing infectious material is dumped directly into a drain
connected to a municipal sewer system. From interviews with coronavirus and influenza researchers, the
primary source of liquid waste are the vacuum traps connected to aspirators in biosafety cabinets that are
used to remove wash buffer and cell culture media from plates containing cells. Small volumes of liquids,
from flasks and tubes, are typically autoclaved as a mode of decontamination and do not apply to this
scenario. Interviews also revealed that a significant fraction of the liquid in these vacuum traps is likely to
be PBS or other non-infectious buffers used to wash cells, and thus any infectious material that is
aspirated is likely to be diluted several fold. As a conservative assumption, we assume the liquid to
contain virus at a concentration of 1E5/mL, and presume a typical flask size of 2L that contains 1L of
liquid when dumped, for a total of 1E8 virus units.
Because no sources were located that identified any human influenza or coronavirus infections from
wastewater (even during influenza season when a lot more infectious material than the amount considered
here enters the sewage system), in this scenario, we consider only the infection of waterfowl exposed to
the wastewater, and therefore limit consideration to avian-adapted avian strains of influenza.
Immediately after dumping, residual chlorine in the municipal water system may neutralize some of the
virus, which has been demonstrated with highly purified samples of HPAI H5N1 and other viruses.1142,1143
With HPAI in idealized conditions there is about a three order of magnitude reduction of live virus, so we
assume a two (log10) reduction. Based on discussions with managers of large and small wastewater
facilities, we conservatively estimate that samples do not separate or mix while in transit until after
arriving at the wastewater treatment facility, at which point they enter a one million gallon (3.8 million
liter) primary clarification tank, at which time we assume the sample fully mixes. Taken together with an
initial sample volume of 1L, a conservative estimate of live virus concentration after chlorine reduction is
a dilution of 1E6 virus units for a final concentration of 2.6E-4/mL.
Further processing steps may reduce the concentration more, but birds have been observed swimming in
open tanks from this point on in the wastewater treatment process.1144 Systems like anaerobic digestion
and UV sterilization are effective at inactivating Avian H5N2 virus1145, but these systems are neither
universal nor used year-round so we do not consider them here.
1142 Rice et al., 2007 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2851495/.

1143 Cromeans et al., 2010 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2820971/.
1144 https://youtu.be/tU416enJAes
1145 Lucio-Forster et al. http://online.liebertpub.com/doi/abs/10.1089/ees.2006.23.897?journalCode=ees

If the city uses a combined sewer system, it may overflow during a rainstorm. Fong et al found no
significant difference in human adenovirus concentration of raw sewage and CSO overflow in
Michigan,1146 which is consistent with other studies of non-pathogenic contaminants yielding overlapping
values.1147,1148 Hence, if the live virus were poured down the drain during heavy rain, the initial sample
would still roughly be diluted into 1M gallons after the initial two (log10) residual chlorine reduction.
Presuming an extremely infectious avian influenza strain with an ingested ID50 of 1 virus unit, and a
conservative assumption that a duck drinks 10mL of water while on the tank, the duck would be dosed
with 2.6E-3 virus units and would be infected approximately with a probability of 2.6e-3 per incident.
However, the frequency that this event occurs is also low. In order for untreated liquid waste to be
dumped, two errors have to occur: the worker who last emptied the flask would have to not put
disinfectant into the flask when returning it to the BSC, and the worker disposing it would have to not put
additional disinfectant in it prior to dumping it. If both of these errors are rules errors with a median
probability of 5e-3, and the flask is conservatively estimated to be emptied once per week, or 50 times per
year, then the overall median frequency of incidents is (5e-3)*(5e-3)*(50) or 1.25e-3/year per lab. Given
that event frequency and the previously calculated probability, the expected frequency of liquid-waste
caused avian influenza infections is (2.6e-3)*(1.25e-3) or 3.25e-6/year per lab (three times per million
years). Despite this estimation making a number of conservative assumptions, this scenario still occurs at
a frequency several orders of magnitude lower than other significant contributors to risk of avian
influenza, and the scenario was not included in the Fault Tree Analysis.
14.1.2.2 Pipe Leak/Burst
In this scenario, untreated liquid waste containing infectious material is dumped directly into a drain
connected to a pipe that is either leaking or has burst, creating a spill out of containment inside the
laboratory building. Leaks within the municipal sewer system, outside the building where the laboratory
is located, are not considered here due to the differences in the dilution, ground filtration, and clean-up
procedures compared to fixing an interior plumbing leak. Municipal sewer leaks would be considered
under the liquid waste disposal scenario, but as mentioned in that section, no human infections of
influenza or coronaviruses caused by wastewater have been reported.
In order for an exposure due to a leaking pipe to occur, two rare events must coincide: the pipe must be
leaking and a laboratory worker must dump infectious liquid waste down the drain. Conservatively using
the maximum pipe failure rate across all pipe sizes and failure types given in a report by the Health and
Safety Executive of the United Kingdom,1149 1E-5 per meter per year, and a conservative expected
maximum of 50m of pipe between the laboratory and municipal sewer system, the maximum expected
failure rate is 5E-4/year per lab. As in the liquid waste disposal scenario, the double errors required for
infectious liquid waste to enter the drain limits the median expected rate of incidents to 1.25E-3/year per
lab. If the pipe leak is conservatively presumed to persist for one-fiftieth of one year prior to being fixed
(approximately seven days), then the overall rate at which liquid waste dumping and pipe leaking
incidents coincide is (1.25E-3)* (5E-4)*(1/50) or 1.25E-8 per year. As spills within the lab occur multiple
orders of magnitude more frequently and are likely involve more concentrated virus, the pipe leak
scenario was neither a significant contributor to risk nor considered further.
1146 Fong et al. 2010. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2813034/

1147 Metcalf and Eddy, Wastewater Engineering, treatment and reuse, 4th ed.
1148 Sztruhár et al 2002, http://www.sciencedirect.com/science/article/pii/S1462075802000080
1149 HSE UK. Failure Rate and Event Data for use within Risk Assessments. http://www.hse.gov.uk/landuseplanning/failure-
rates.pdf. Last Update June 28th 2012. Accessed November 23rd 2015.

14.1.3 The Monte Carlo framework
Monte Carlo simulations were performed for each event tree to obtain probability estimates for each
potential outcome, the estimated frequency of each outcome, estimated reduction factors, and the amount
of material released associated with each outcome.
As discussed in more detail in Sec. 6.2.5, each event tree consists of a series of potentially conditional
nodes, each of which represents a step in the process at which an error or failure could potentially occur.
Each node can result in either a “success” or “failure” (yes/no) outcome, and each success/failure
outcome potentially affects which subsequent nodes are relevant and, in some cases, the probabilities
associated with the success/failure likelihoods for subsequent nodes. In addition, for some nodes, either a
success or a failure could result in a reduction in the amount of potentially infectious material available
for release. In order to reflect both the statistical or probabilistic uncertainty as well as the uncertainty
associated with the true value of a parameter (due to epistemic or aleatory uncertainty), probability
distributions were assigned for key input parameters for each event tree (e.g. number of opportunities,
amount of material being handled) and for each node within each tree (e.g., probability of failure,
reduction factors, etc.), and Monte Carlo simulations were performed.
For each fault tree, a probability distribution was assigned for the following factors:
• Number of opportunities per year for the event to occur and
• Amount of potentially infectious material available for release during any given opportunity (i.e.,
how much material at what concentration is being handled at any given point in time). In most
cases, the amount available for release was defined as the product of two independent random
variables: 1) the volume of material being handled and 2) the concentration (viral titer) of the
material.
For each node in each tree, either a probability distribution or a fixed value was assigned for the following
factors:
• Probability of success/failure (in some cases, this probability distribution was dependent on the
outcome of previous nodes and/or the volume of material being handled), and
• Reduction factor (the fraction of infectious material is removed from the material potentially
released) if a success or failure is realized (in some cases, these probability distributions were
dependent on the outcome of previous nodes and/or the volume of material being handled).
For a given tree, the Monte Carlo simulation was performed by generating 2.5 million random outcome
realizations. For each of the 2.5 million realizations, the following steps were performed:
1. Generated a random realization of the amount of potentially infectious material being handled,
called the “Material Available for Release” (MAR).
2. For each node in the tree, assigned a probability of failure (pfi, where i is the ith node), either as a
random realization from the distribution of possible probabilities for that node, or a fixed value if
no distribution was defined. For conditional nodes (i.e., nodes with success/failure probabilities
that are dependent on the result of previous nodes or the volume of material being handles), the
assignment of probabilities took these results into account.

3. For each node in the tree, based on the assigned probability of failure, generated a random
success or failure outcome (where the probability of success is 1- pfi and the probability of failure
is pfi).
4. For each node in the tree, based on the realized success or failure outcome, determined the
reduction factor (amount by which the amount of material being handled is reduced by the
success of failure of that node). The reduction factor for a given realization was generated either
as a random realization from the distribution of possible reduction factors associated with a
success or failure outcome for that node, or a fixed value if no distribution was assigned. Note
that for some trees, different reduction factors were assigned based on the type of potential
exposure (e.g., fomite or aerosol exposure).
5. Based on the series of successes and failures for a given realization, determined whether an
exposure occurred and, if so, the type of exposure that occurred (e.g., personal aerosol exposure,
hand fomite exposure, subcutaneous exposure, etc.). This determination was based on the
description of each tree.
6. For each realization that resulted in an exposure, computed the overall reduction factor as the
product of reduction factors realized for each node, as well as the mass of potentially infectious
material involved in the exposure. The mass of material involved in the exposure (“Q”) was
computed as the MAR for a given realization multiplied by the overall reduction factor for that
realization. Note that for some trees, overall reduction factors and resulting Q values were
computed separately for different types of exposures (e.g., fomite or aerosol exposure).
All results from every one of the 2.5 million passes through the tree were stored, allowing various
summary statistics to be computed. For each node, the observed proportion (estimated probability) of
failure, the average and standard deviations for the observed (realized) reduction factors when the node
was successful and the average and the standard deviations for the observed reduction factors when the
node was a failure was computed. Note that reduction factor averages and standard deviations were
computed for all relevant types of exposure (e.g., aerosol, fomite).
For each unique “trace” through the tree (i.e., each unique pattern of successes and failures), the observed
proportion of the 2.5 million runs that resulted in that unique trace, representing the estimated probability
of that trace, was computed. In order to more fully understand the uncertainty associated with the
estimated probability for the trace, an additional set of calculations was performed for each trace, wherein
the probabilities that had been assigned to each node for a given pass through the tree (i.e., the pfi values)
were used to compute the probability of the trace (i.e., the probability of that trace’s unique pattern of
successes and failures) for each of the 2.5 million passes through the tree. This approach generated a
distribution of 2.5 million probabilities per trace. The average and standard deviation as well as the 1st,
5th, 50th, 95th, and 99th percentiles were computed. Also, the averages and standard deviations of the
overall reduction factors associated with the trace were computed. Note that reduction factor averages
and standard deviations were computed for all relevant exposure types. Lastly, the averages and standard
deviations of the Q values (the product of MARs and the reduction factors) associated with the trace were
computed. Q value averages and standard deviations were computed for all relevant exposure types.
Because numerous traces resulted in the same “exposure outcome” (e.g., different series of successes and
failures could all lead to a hand fomite exposure), an additional summary table was created that
summarized the results across all traces associated with an exposure outcome. For each unique exposure
outcome, the following statistics were computed and summarized:

• Observed proportion of the 2.5 million runs that resulted in the exposure outcome, which
represents the estimated probability of the exposure outcome and was computed as the sum of the
probabilities (proportion of observed occurrences) for all unique traces that resulted in the
exposure outcome. For example, if three different traces all resulted in an environmental aerosol
exposure, the probability of an environmental aerosol exposure was computed as the sum of the
probabilities for each of those three traces.
o Similarly, the uncertainty in the estimated probability for a given exposure outcome was
captured by computing the probability of that outcome for each of the 2.5 million passes
through the tree. For a given pass, the probability of the exposure outcome was computed as
the sum of the probabilities for each trace that resulted in the given exposure outcome, which
generated a distribution of 2.5 million probabilities for the exposure outcome. The average
and standard deviation as well as the 1st, 5th, 50th, 95th, and 99th percentiles were then
computed for this probability distribution.
• The averages and standard deviations of the overall reduction factors associated with the
exposure outcome (for all relevant exposure types).
• The averages and standard deviations of the Q values associated with the exposure outcome (for
all relevant exposure types).
• The range of potential frequencies of occurrence for each exposure outcome was also computed.
The first step was to generate 2.5 million realizations of the number of opportunities, based on the
probability distribution for the number of opportunities per year for the event tree. The product
between the ith opportunity count and the ith probability of the exposure outcome was then
calculated for each of the 2.5 million opportunity counts and probabilities, to generate 2.5 million
“expected frequencies”. The average and standard deviation as well as the 1st, 5th, 50th, 95th, and
99th percentiles of these frequencies were then computed.
14.1.4 Human Reliability Assessment in Biological Laboratories
Many years of continuous improvement in containment laboratory design has reduced the failure of the
many mechanical containment features to rates below that of human reliability. A significant fraction of
this reduction in failure rate comes from the mechanical redundancies, interlocks, and alarm systems that
require a cascading series of improbable events to occur prior to a loss of containment event. In addition,
the interlocks and alarms provide a visual, auditory, or physical alert that a failure has occurred,
converting previously covert failures into overt ones, and allowing workers present in the facility to cease
work and rectify the failure or error condition prior to a loss of containment event.
In contrast to these mechanical failures, human errors often remain covert, and a single human error can
inadvertently subvert many mechanical or physical safety features simultaneously. For example, while a
typical pass-through autoclave used in a BSL3 facility may contain a temperature readout, pre-
programmed cycles to ensure proper time, alarms that report failure conditions, and a physical interlock
that prevents the clean side doors from opening unless a complete, successful cycle has finished, an
operator that does nothing more than overload the autoclave due to naiveté or momentary forgetfulness
can result in still-contaminated material leaving the containment suite. In addition to human errors
subverting safety features, due to the design of the mechanical safety features, many loss-of-containment
scenarios are unlikely to occur unless precipitated via human error.

Finally, human errors can exacerbate a mechanical failure or loss of containment event. For example,
workers who misinterpret, ignore, or otherwise silence alarms, whether caused by misbehavior or
ignorance, convert a routine response to a mechanical failure into a potential covert loss of containment
event. Workers who experience potential exposures and ignore the established response protocol due to a
self-assessed belief that the risk of the exposure is low, or, conversely, a fear of shame and consequences
should the incident be reported, can increase the chance that an exposure leads a laboratory-acquired
infection. Moreover, should a worker fail to follow an isolation protocol after an exposure, a laboratory-
acquired infection may initiate a local outbreak.
For these reasons, assessing human reliability in containment labs is a critical component to modeling the
risk of loss-of-containment events. Although some human error rates are available for specific types of
biological laboratory accidents, no comprehensive Human Reliability Analysis (HRA) study has yet been
completed for a biological laboratory. Assigning approximate human error probabilities for specific event
nodes in the model accident trees required finding suitable proxies for accidents in other fields.
Operations research conducted to facilitate the assignment of data from one context for human errors to a
another has shown that, as a first approximation, human errors can be grouped into a few generic error
categories by behavior type, with associated “rule of thumb” accident error ranges.1150,1151,1152 These
values are then typically refined by researchers, for example based on employee error rates on equipment
simulators used during training or surveyed rates during operation.1153
The classification system used here is derived from nuclear power plant HRA studies, and consists of
three categories: rule-based, skill-based, and knowledge-based errors.1154,1155 Table 15.1 below
summarizes the classification system adapted for use in this study.
1150 In addition to unstructured searches for operations research literature, a systematic search for all sources mentioned in the
bibliography of a recent textbook on HRA studies was conducted: Anthony J. Spurgin, Human Reliability Assessment:
Theory and Practice (Taylor & Francis Group, 2009).
1151 See for instance: Charles P. Shelton, “Human Interface/Human Error,” 18-849b Dependable Embedded Systems, Spring
1999, http://users.ece.cmu.edu/~koopman/des_s99/human/. Accessed August 3, 2015. Based on data from Barry Kirawn, A
Guide to Practical Human Reliability Assessment (London: Taylor and Francis Ltd., 1994).
1152 Other factors, such as the amount of time available to rectify a mistake before an accident occurs, can then be incorporated
as adjustment factors. See for example: Ronald L. Boring, David I. Gertman, “Human Error and Available Time in SPAR-
H,” CHI 2004 Workshop on Temporal Aspects of Work for HCI,” p. 3, http://www.aeso.ca/downloads/2009-02-
06_Study_of_Human_Error_rates.pdf. Accessed August 3, 2015.
1153 Pierre Le Bot, “Human reliability data, human error and accident models – illustration through the Three Mile Island
accident analysis,” Reliability Engineering and System Safety 83 (2004): p. 154.
1154 Electric Power Research Institute (EPRI), “Systematic Human Action Reliability Procedure (SHARP),” EPRI NP-3583,
Project 2170-3, Interim Report, June 1984, A-8, http://www.epri.com/abstracts/Pages/ProductAbstract.aspx?ProductId=NP-
3583. Accessed August 3, 2015.
1155 The four-tier categorization process used in the following source was also consulted to define cases: Scott Shappell, Doug
Wiegmann, HFACS Analysis of Military and Civilian Aviation Accidents: A North American Comparison, ISASI 2004
http://www.asasi.org/papers/2004/Shappell%20et%20al_HFACS_ISASI04.pdf.

Table 14.1. General human error types as applied to a biological laboratory
Human Definition Error rate Accident Examples

error improves with probability
type1156 range
Redundant
Errors in following 5E-4 to 5E-2, Omitting a required
checking; Written
Rule-based instructions or set procedures, log uniformly PPE item, violating
rules vs. oral
accidentally or purposefully distributed isolation
instructions1157
Cutting oneself with a
Redundant 5E-5 to 5E-3,
Errors involving motor skills sharp object, creating
Skill-based processes; log uniformly
involving little thought a splash while
Practice distributed
pipetting
Identifying an
Errors stemming from a lack incorrectly labeled
5E-3 to 5E-1,
Knowledge- of knowledge or a wrong package as actually
Experience log uniformly
based judgement call made based on hazardous, choosing
distributed
a lack of experience the proper centrifuge
tube
A search for human error data in other fields was conducted to verify the validity of the chosen error
ranges and to refine range estimates in specific cases. Data sets and associated reports on human errors in
1157 Based on numbers and discussion in: A. D. Swain, H. E. Guttmann, “Handbook of Human Reliability Analysis with
Emphasis on Nuclear Power Plant Applications: Final Report,” NUREG/CR-1278, SAND80-0200, August 1983,
http://pbadupws.nrc.gov/docs/ML0712/ML071210299.pdf. Accessed August 3, 2015.

the nuclear,1158,1159,1160 aerospace,1161 aviation,1162,1163,1164,1165 medical,1166,1167,1168,1169 and hazardous
materials sectors,1170 in the workplace,1171,1172 and with motor vehicles,1173 were compiled and reviewed.
Extracting useable human error rates from the available accident data requires knowing the total number
of accidents caused by human errors (the numerator) and the total number of operations that could have
led to an accident (the denominator). In general, human error data suffers from a lack of values for the
denominator. Accidents per year are often tallied in the literature, and the number of said accidents
attributable to human error are sometimes available, but very few studies can provide a count of the total
1158 A. D. Swain, H. E. Guttmann, “Handbook of Human Reliability Analysis with Emphasis on Nuclear Power Plant
Applications: Final Report,” NUREG/CR-1278, SAND80-0200, August 1983, p. 15-14, p. 6-17, p. 20-38, p. 15-5,
http://pbadupws.nrc.gov/docs/ML0712/ML071210299.pdf. Accessed August 3, 2015.
1160 M. K. Comer, D. A. Seaver, W. G. Stillwell, C. D. Gaddy, “Generating Human Reliability Estimates Using Expert
Judgement Volume 2. Appendices,” NUREG/CR-3688/2 of 2, SAND84-7115, November 1984, p. C-1 – C-10,
http://prod.sandia.gov/techlib/access-control.cgi/1984/847115-2.pdf. Accessed August 3, 2015.
1161 Chandler F, et. al. (2010) “NASA Human Error Analysis,” http://www.hq.nasa.gov/office/codeq/rm/docs/hra.pdf. Accessed
August 3, 2015.
1162 Garibay A, Young J (2013) “Reducing General Aviation Accidents By Utilizing Airline Operational Strategies,” Aviation
Technology Graduate Student Publications, Paper 25: 6,
http://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=1019&context=atgrads. Accessed July 1, 2015.
1163 Shappell S (2006) “Human Error and Commercial Aviation Accidents: A Comprehensive, Fine-Grained Analysis Using
HFACS,” http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA463865. Accessed July 1, 2015.
1164 Maurino D (2000) “Human Factors and Safety Management: The Role of the Regulator,” Flight Safety and Human Factors
– ICAO, 14th Annual FAA/CAA/TC Human Factors in Aviation Maintenance Symposium, Vancouver, Canada.
http://www.faa.gov/about/initiatives/maintenance_hf/library/documents/media/mx_faa_%28formerly_hfskyway%29/14th_s
ymposium/human_factors_and_safety_management_the_role_of_the_regulator.pdf. Accessed July 1, 2015.
1165 Shappell S, Wiegmann D (2004) “HFACS Analysis of Military and Civilian Aviation Accidents: A North American
Comparison,” Australian Society of Air Safety Investigators ISASI,
http://www.asasi.org/papers/2004/Shappell%20et%20al_HFACS_ISASI04.pdf. Accessed July 1, 2015.
1166 Committee on Quality of Health Care in America, Institute of Medicine, To Err is Human, eds. Linda T. Kohn, Janet M.
Corrigan, Molla S. Donaldson (Washington: National Academies Press), p.1, 28, 31-34.
1167 Marx D (2001) “Patient Safety and the “Just Culture”: A Primer for Health Care Executives,”
http://www.safer.healthcare.ucla.edu/safer/archive/ahrq/FinalPrimerDoc.pdf. Accessed July 1, 2015.
1168 Centers for Disease Control and Prevention, “Inpatient Surgery,” April 29, 2015, http://www.cdc.gov/nchs/fastats/inpatient-
surgery.htm. Accessed July 1, 2015.
1169 Benger JR, Lyburn ID (2003) “What is the effect of reporting all emergency department radiographs?,” Emergency
Medicine Journal: 40-43, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1726029/pdf/v020p00040.pdf. Accessed August 3,
2015.
1170 U.S. Department of Transportation, Pipeline and Hazardous Materials Safety Administration, “Top Consequence 2005-
2009: Hazardous Materials by Commodities & Failure Modes,” Issue 3, September 1, 2011, p. 9,
http://www.phmsa.dot.gov/pv_obj_cache/pv_obj_id_3340E5EE847704EA2F3FBF73F59757A324780700/filename/Top%2
0Consequence%20Hazardous%20Materials%20Commodities%20Report.pdf. Accessed July 1, 2015.
1171 United States Department of Labor, Bureau of Labor Statistics, “Occupational Injuries/Illnesses and Fatal Injuries Profiles,”
data retrieved at http://data.bls.gov/gqt/InitialPage on July 16, 2015. The datasets are split between non-fatal and fatal
injuries. For non-fatal injuries, select either "Case and Demographic Numbers" or "Case and Demographic Incidence Rates".
Then under "Characteristic type", select either: "Source of injury/illness" to get the equipment or harmful substance leading
to the accident (ex. acids), or "Event or exposure" to get the type of incident that occurred (ex. bitten and struck by animal).
The event types studied were: “Bitten and struck by animal,” “exposure intact skin, eyes, or other exposed tissues,”
“exposure scratch or other open wound,” “exposure through medical injection,” “exposure unintentional needlestick, sharp
injury,” and “needlestick without harmful substance.” The work sectors considered were “all,” “health care and technical,”
and “computer, engineering, and science.” For fatal injuries, the datasets extracted were fatal occupational injuries for
biological scientists (code 19102x).
1172 Brown A, Patterson D (2001) “To Err is Human,” Proceedings of the First Workshop on Evaluating and Architecting
System Dependability http://roc.cs.berkeley.edu/papers/easy01.pdf. Accessed August 3, 2015.
1173 National Highway Traffic Safety Administration (NHTSA), “An Examination of Driver Distraction as Recorded in NHTSA
Databases,” Traffic Safety Facts: Research Note, p.1, http://www-nrd.nhtsa.dot.gov/Pubs/811216.pdf. Accessed July 1,
2015.

number of procedures per year that could have led to accidents, so a rate is impossible to obtain.
Gathering denominator data is difficult and expensive, often requiring direct observation. For instance,
biological research workplace accident data by accident type retrieved from the Bureau of Labor Statistics
could not be used, in part because it was impossible to determine how many actions were undertaken per
year that could have led to injuries such as “falls, slips, [or] trips.”1174 For several sectors–such as
workplace accidents, motor vehicle accidents, and hazardous shipment accidents– HRA studies are rarely
conducted because insurers and regulators are primarily interested in determining risk group and risk
factors based on number and severity of accidents per year.
Sources with a large sample of total accidents, such as the aforementioned Bureau of Labor Statistics
dataset, also often lacked the level of granularity in accident types needed to ensure accident situations
were analogous to plausible incidents in high-containment laboratories. Even when considering the list of
accidents relevant to biology, this comparison would be dubious. For example, a marine biologist slipping
and injuring themselves while working along the shore would have been categorized as a “biologist –
falls, slips, trips” event, but this situation bears little resemblance to the analogously categorized biologist
slipping and injuring themselves while working in a high-containment laboratory. Since the dataset
categories compressed a wide range of accident types and accident variants together, it was not possible
to combine the data from this large dataset with human error studies with small sample sizes to obtain
trustworthy accident rates applicable to high-containment laboratory work.
In other fields, such as surgical medicine and especially commercial and military aviation, routine
operation involves an extremely large number of individual actions and non-actions, complicating the
extraction of useable denominator values. To take an example from commercial aviation, roughly two
errors are committed per flight, the vast majority of which have no consequences and are not noticed by
the flight crew.1175 In addition, accidents in these contexts are often complex situations caused by a
combination of human and mechanical failures exacerbated by abnormal operating conditions.1176 In such
cases, the number of incidents solely attributable to human error is difficult to obtain. As a result, human
error rates in potential proxy fields often could not be reliably determined.
Notable exceptions were found in reports for the nuclear and aerospace industries, where the potential
catastrophic consequences of errors has motivated detailed HRA research. The human error data used in
this section was mostly derived from these sources, principally from the Electric Power Research Institute
(EPRI) “Systematic Human Action Reliability Procedure (SHARP)” document NP-3583.1177 These were
complemented by values taken from the Sandia Laboratories NUREG/CR-1278 “Handbook of Human
Reliability Analysis with Emphasis on Nuclear Power Plant Applications”.1178 More specifically, the
sections of the NUREG/CR-1278 used were those on omitting steps listed out on written instructions,
1174 United States Department of Labor, Bureau of Labor Statistics, “Occupational Injuries/Illnesses and Fatal Injuries Profiles,”
data retrieved at http://data.bls.gov/gqt/InitialPage on July 16, 2015.
1175 Maurino D (2000) “Human Factors and Safety Management: The Role of the Regulator,” Flight Safety and Human Factors
– ICAO, 14th Annual FAA/CAA/TC Human Factors in Aviation Maintenance Symposium, Vancouver, Canada.
http://www.faa.gov/about/initiatives/maintenance_hf/library/documents/media/mx_faa_%28formerly_hfskyway%29/14th_s
ymposium/human_factors_and_safety_management_the_role_of_the_regulator.pdf. Accessed July 1, 2015.
1176 See for instance data in Table 2 of: Scott Shappell, “Human Error and Commercial Aviation Accidents: A Comprehensive,
Fine-Grained Analysis Using HFACS,” July 2006, p. 7, http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA463865.
1178 Swain AD, Guttmann AE (1983) “Handbook of Human Reliability Analysis with Emphasis on Nuclear Power Plant
Applications: Final Report,” NUREG/CR-1278, SAND80-0200 http://pbadupws.nrc.gov/docs/ML0712/ML071210299.pdf.
Accessed August 3, 2015.

misremembering oral instructions, misreading labels, and detecting errors.1179
Three general types of errors were most frequently used in the analysis. The first of these was the rule
error, which were errors incurred in any laboratory task where a prescribed procedure or rule applied to a
task, as in, for example, wearing PPE, including safety features in a centrifuge, or washing one’s hands
when leaving a laboratory. For general rule errors where no specific cause of the failure could be
assigned, the entire failure rate range listed in the source, 5E-4 to 5E-2 per attempt, was used, distributed
log normally (i.e., uniformly distributed on the exponent). When a specific type of rules error failure with
an assignable probability was believed to be the most likely cause of the error, it was used as the mode of
a log triangular distribution over the range. This parameter range was used, for example, in the
application of an error occurring while following a protocol of more than ten steps to the failure to
properly package a shipment of infectious material. Rules errors include failures to follow rules due to
any cause, including ignorance, forgetfulness, or willful disobedience.1180
The second general error used was the skill-based error, which are errors involving motor skills, in, for
example, handling a sharp instrument during necropsy. In order for this type of error to apply, the task
must be one where the motor skill of the individual would improve over time with practice. For these
errors, the entire failure rate range listed in the source, 5E-5 to 5E-3 per attempt, distributed log normally,
was used. Skill errors were not assigned to basic motor tasks a worker would also attempt outside of a
laboratory, such as holding an object without dropping it, or walking without tripping, as these are not
motor tasks for which the failure rate would likely decrease with worker practice.
The third general category of error was the knowledge error, which were errors caused by intellectual
naivety or misunderstanding, and are a type of error whose probability decreases through experience with
the topic. For example, PAPR failures, such a disconnected tube or low battery, would be self-announcing
via a silent fan and reduced airflow. Workers with extensive experience with PAPRs would be more
likely to immediately notice the change in sensation versus a worker who had just begun using PAPRs.
Like the rules error, for general knowledge errors where no specific cause could be assigned, the entire
failure rate range from the source, 5E-3 to 5E-1 applied and was log normally distributed. When a
specific cause could be identified, that probability was assigned to the mode of a log triangular
distribution over the range. In certain cases, the upper limit of the range was restricted to a lower value of
1E-1, reasoning that only the persons least experienced with the task would fail at the original limit of 5E-
1, and interviews with practicing influenza and coronavirus researchers repeatedly revealed that all
workers in the high containment laboratories had practice and training in lower containment before
entering the laboratory.
In addition to these three general ranges, which applied to the majority of the errors appearing in the event
trees, two other errors were used. The first was the previously mentioned failure to follow a protocol of
greater than ten steps, which applied when an error was committed in attempting a task for which a
detailed protocol is likely to be present. The second was a failure due to misreading an label, which
applied, for example, when a label on a package was misunderstood and resulted in the package being
mis-delivered.
It should be noted that these probabilities, as applied, are the best, but still flawed, approximation of the
risk of accidents or errors as committed in a biological laboratory. Biological researchers tend to be
highly skilled with many years of experience prior to entering a high containment space, which could lead
1179 Ibid.
1180 In the biosafety section, intentional violations of the rules are assumed to be committed wihtout malicious intent, but instead
were due to laziness or the false assumption by the worker that the risk of accident was negligible enough not to bother with
the required procedure or equipment, a type of violation seen in historical accident reports and mentioned by interviewees.

to errors being committed at rates near the lower limit of the ranges used here. While the ranges used have
been carefully calibrated to represent a general range of errors of this type across industries, possibly,
biological researchers commit errors at frequencies outside the range given due to their level of training
and education. Any primary research into the types of errors and their frequencies committed in biological
laboratories has the possibility to increase both the precision and accuracy of the error rates incorporated
into the fault tree analysis in this report, reducing uncertainty. Given the significance of human error in
driving risk, and the possible consequence of an accident in a containment laboratory, such a study would
likely have great utility.
14.1.5 Further Information on Modeling Infection Risk Caused by Fomites
14.1.5.1 Summary
A stochastic Markov model was developed to predict the likelihood of an outbreak initiating after a
laboratory worker leaves containment with virus on his or her person. The model tracks the contamination
through the paths it must take to result in infection of the initial laboratorian, of one or more household or
community members, or of avian species on a commercial farm (or any combination of the three) (Figure
14.1). All infections are the result of internalization of the virus from a contaminated surface or body part;
that is, this is a model of contamination transference and subsequent infection, not a model of contagious
transmission. Any avian infections resulting from this model are assumed to spread throughout the flock
and cause large outbreaks on the scale of historical avian influenza outbreaks. Epidemiological spread of
human infections is modeled in the branching process model for local outbreaks and the IIM SEIR model
for global outbreaks.
The transference model utilizes Monte Carlo simulations that string together the likelihoods of a number
of possible actions that would lead to internalization, spread, or removal of the virus (namely, contacting
your eye, nose, or mouth; physical contact with another person; contacting surfaces and fomites through
regular activity; handwashing; and showering). The frequency of each of these events is described as a
rate per minute, and for each minute of model time a random draw from a binomial distribution (with a
probability equal to the event rate) determines if the event occurs. The viability of the virus also decreases
according to its half-life on skin or nonporous surfaces over the course of the model time. Each contact
event, whether to a fomite, surface, or person, transfers a certain fraction of the virus, based on data
collected from published studies on transfer of viral material. Human infection occurs when viral
contamination on a person’s hand enters their mucosal membranes of the eye, nose, or mouth, and is
dose-dependent based on the calculated amount of virus present at the time of inoculation.
For an animal infection to occur, the primary laboratorian must visit a farm housing a susceptible species,
at which point it is assumed that all of the virus is inoculated into the animal. For animal contact to occur,
the worker may need to violate quarantine protocols, which occurs at a specific probability, after which
visits to an animal facility occur at a predetermined rate, as with the events above.

Figure 14.1. Schematic of the transference and infection model.
14.1.5.2 Model Structure
The transference model simulates three types of events: spreading of contamination from an individual’s
hands to another person or animal; loss of virus onto surfaces through contact, washing, or virus
degradation; and inoculation of a susceptible species, either one’s self (through inoculation of mucosal
membranes in the eye, nose, or mouth from contamination on an individual’s hands) or an avian species
at a poultry farm. In every unit of model time, each of these events may or may not occur, based on a
specified rate of occurrence. Through repeated such events, the model predicts whether the laboratorian
causing the loss of containment is infected, how many other people in the laboratorian’s household or the
community are infected, and if any avian species are infected (in the case of avian influenza). These
calculations are performed for a number of simulated releases, allowing determination of a frequency of
each consequence occurring.
The model functions by evaluating the likelihood of each event happening during every minute from the
initial release (i.e., exit of containment by a worker carrying contamination) through the next 24 hours.
The likelihood of each event occurring in any given minute is determined by the average rate of
occurrence, and is independent of other events happening. Whether a given event occurs within a given
minute is determined by a random draw from a binomial distribution with probability p = events per
minute. Minutes were chosen as the unit of model time so that all event rates would be less than one.
Model events occur sequentially, that is, two events cannot happen simultaneously, even within a given
minute. The events are therefore evaluated in a predetermined order within each minute modeled, as
follows:
1. Worker touches his or her eye, nose, or mouth

2. Worker touches household member
3. Worker touches community member
4. Worker touches surfaces
5. Worker washes hands
6. Worker showers
7. Remaining virus is fractionally degraded

Each of the possible events is described in further detail below.
14.1.5.2.1 Spread of Virus
Each model simulation begins with an amount of virus (in PFU) contaminating a laboratorian’s hand as
he or she leaves the containment facility. This viral contamination can be spread to other individuals in
the worker’s household or in the community at large. The spread of virus is calculated as follows. For
every minute modeled, the worker may come into contact with a family member and may also contact a
member of the community. Whether either of those contacts is made is based on specified frequencies of
occurrence for each type of contact. A random draw from a binomial distribution determines whether the
contact happens in each step of the model time.
Contact with other individuals is assumed to be through hand-to-hand contact (e.g., handshakes), as that is
a common form of personal contact and the type most likely to cause subsequent inoculation through
touching of one’s face (it is more likely someone will touch their face with a contaminated hand than a
contaminated forearm or shoulder). When contact occurs, a fraction of the virus is transferred to the
recipient’s hand. The initial worker can spread viral material to several other individuals through multiple
contact events over the course of the model run (within each minute modeled, however, the worker can
only touch at most one household member and one community member). Further spread of the virus from
those contacted to additional generations of recipients is not followed by the model.
14.1.5.2.2 Loss of Virus
In addition to spreading of contamination from person to person, viral material can be lost through
touching inanimate objects and surfaces, washing hands and showering, and through the natural decay
that occurs in viruses on surfaces. Therefore, even without contacting other individuals, the likelihood of
a contaminated laboratorian infecting himself or herself will diminish over time as these events occur.
As with the contact events, viral loss through touching surfaces, handwashing, and showering occurs
based on specified rates. Each event has a separate rate, and for each minute modeled a draw from a
binomial distribution determines whether the event happens. Virus loss events occur within a single unit
time in the following order: contact of surfaces, handwashing, showering; all loss of virus events occur
after spreading events (described above).
Degradation of the virus occurs as the final event of every unit of model time (unlike the previously
described events, which may or may not occur). The amount of virus remaining is calculated as an
exponential decay function based on the amount of virus at the end of the unit of model time (after all
contact and washing events) and the half-life of the virus on skin, as demonstrated in the following
equation:
𝑁𝑡 = 𝑁0 2−𝑡/𝑡1/2
Where:
t = Duration of time passed (i.e., one minute)
t1/2 = Half-life of virus on skin (in minutes).
N0 = Amount of virus before degradation
Nt = Amount of virus remaining after degradation

14.1.5.2.3 Human Infection
The initial laboratorian can become infected by inoculating himself or herself in the mucous membranes
of the eye, nose, or mouth. As with other contact events, in each minute modeled the worker may touch
his or her face, depending on a random draw from a binomial distribution based on a specified rate of face
touching. When a person touches his or her face, a fraction of the virus is transferred to the mucous
membrane (the same fraction as is transferred during hand-to-hand contact). Virus accumulates in the
body with each face contact event, and all face locations (eye, nose, and mouth) are considered as one. At
the end of the modeling run, the total amount of virus accumulated in the body is used to determine the
probability of infection, based on a probit dose-response function.
Secondary recipients of the laboratorian’s contamination also become infected through self-inoculation
following face touching events. However, instead of modeling each additional contaminated person
completely, a single mock individual is modeled and used to determine the amount of virus remaining on
a secondary person’s hand at any point in time. For each secondarily contaminated individual, the time
until the first face contact event is determined by a random draw from an exponential distribution with
parameter λ = 1/rate of contact.1181 The fraction of virus remaining on the mock individual at this time
post-contamination is then used to calculate the amount of virus remaining on the contaminated
individual, based on the amount that was transferred during the hand-to-hand contact event. A fraction of
the virus on the individual’s hand is internalized (the same fraction that is transferred during hand-to-hand
contact) and the probability of infection is calculated based on a probit dose-response function. The
calculation of the internalized dose for secondary recipients of contamination ignores subsequent face-
touching events; however, in most cases the first contact event contributes the vast majority of
internalized virus, and the contribution to total dose of subsequent events is negligible.
14.1.5.3 Animal Infection
The initial laboratorian can infect animals if he or she visits a farm location housing susceptible species.
Only farmed poultry species are considered in the model, as contacts between people and wild birds are
exceedingly rare. The model estimates whether an outbreak in birds is initiated only, not the size of the
outbreak.
Only a certain portion of laboratory workers will ever contact a susceptible avian species. Workers who
do contact susceptible species do so at a specified rate. For each minute modeled, whether or not animals
are contacted by the worker is evaluated by a random draw from a binomial distribution. Additionally,
staff working with avian influenza are required to follow a five-day quarantine period1182 wherein they
cannot contact any avian species, which is modeled with a specified rate of failure to adhere to protocol.
When a worker does contact an animal, all virus on the worker’s hand at the time of contact is assumed to
be inoculated into the animal, and the probability of infection of the animal is determined by a probit
dose-response function. Animal contact is the last event evaluated within each minute modeled, and thus
the amount of virus transferred to the animal is the amount remaining on the worker’s hand after all prior
contact, washing, and degradation events.
1181 For any event that happens randomly at an average rate r, the durations of periods between each event follow an exponential
distribution with rate parameter λ = 1/r.
1182 Centers for Disease Control and Prevention, (2013c) Interim Risk Assessment and Biosafety Level Recommendations for
Working With Influenza A(H7N9) Viruses.

14.1.6 Sources used to identify incident scenarios to include in the study
In this study, the following previous laboratory accident risk assessments were analyzed:
• National Bio- and Agrodefense Facility (NBAF) Updated Site-Specific Biosafety and Biosecurity
Mitigation Risk Assessment (DHS 2012),
• NBAF Site-Specific Biosafety and Biosecurity Mitigation Risk Assessment (DHS 2010),
• BioSquare Phase II (NEIDL) Supplemental Final Environmental Impact Report (Boston U 2013),
• Biological Defense Research Program Final Programmatic EIS (DOD 1989),
• Final Revised Environmental Impact Study (EIS) for the Proposed Construction and Operation of
a BL3 Facility at LLNL (DOE 2008),
• Environmental Assessment for the Proposed Construction and Operation of a BL3 Facility at Los
Alamos National Laboratory (DOE 2002),
• Final EIS, Rocky Mountain Laboratory Integrated Research Facility (NIH 2004),
• Final EIS for George Mason University Biomedical Research Laboratory (NIH 2008),
• Final EIS for University of Louisville Center for Predictive Medicine Biodefense EID Regional
Biocontainment Laboratory (RBL) (NIH 2007),
• EIS for Colorado State RBL (NIH 2003), and
• Evaluation of Health and Safety Risks of the New US Army Medical Research Institute of
Infectious Disease High-Containment Facilities at Fort Detrick [NRC 2010]
Most of these studies were completely qualitative and lacked descriptions of possible loss-of-containment
scenarios. These studies provided some semi-quantitative calculations based on hypothetical scenarios
with notional parameters. For these qualitative studies, the scenarios that were considered were simply
noted for inclusion. The NBAF studies were quantitative because pathways, frequencies and
consequences of loss-of-containment events was explicitly calculated. Events were characterized as high,
medium or low frequency (the consequences were calculated for a non-zoonotic virus, so could not be
used as a direct comparison). The NEIDL study based their probabilities of accidents on historical
incident reports and then assessed risk based on the historical frequency of these accidents and the
estimated number of people potentially exposed. Using their data, we characterized every event as high
risk (causing more than one human exposure per year), medium risk (causing more than 1 human
exposure per 10 years) and low risk (less than one human exposure every 10 years).
These previous reports were supplemented by incident/accident reports, including:
• NIH RDAC list of reported incidents, 1977-April 2015 (NIH),
• CDC Select Agent Reports 2003-2009 (CDC—obtained in the appendix of the NEIDL document,
above), and

• Various publically available BSL3&4 accidents (Various—summarized in NEIDL appendix).
From these reports, an incident was characterized as high risk if it represented 10%% or more of accident
reports, medium risk if it represented 2% or more of accident reports, and low risk if it accounted for less.
14.2 Methodological Details of the Branching Process Model
At the early stages of a nascent outbreak, when a small number of people are infected, stochastic variation
between individuals plays a significant role in the eventual size of the outbreak—whether it extinguishes
at a small number of cases either due to chance or human intervention, or grows to a large epidemic or
global pandemic. For example, if an outbreak begins with a single individual, and this individual self
isolates, perhaps due to symptom severity, no other persons may be infected and the outbreak terminates.
In contrast, as witnessed in the recent outbreak of MERS in South Korea, a single individual that contacts
a large number of people may single-handedly spark a large epidemic. Deterministic models, such as
SEIR models, while appropriate for large epidemics in which the number of infected individuals ensures
that the mean adequately describes the behavior of most individuals, cannot capture the individual
variation inherent in the early stages of an outbreak, and therefore may overestimate the probability that a
loss of containment event spreads beyond local control.
In contrast, models that account for this early stochastic variation are more likely to paint an accurate
picture of the early phases of a disease outbreak. For this report, a branching process model (BPM) was
used in discrete time to capture the individual stochastic variability in nascent outbreaks. Branching
process models simulate a “birthing process” over time, and, in the discrete-time branching process model
used here, time is represented by generations of infection. Each individual at generation g has a
probability of “birthing” (i.e. infecting) a number of offspring (infected individuals) in generation g+1
described by a probability distribution (termed the “offspring distribution”). This birthing process is then
repeated for a specified number of generations or until some desired stopping point or exit condition is
achieved. Typically, the offspring distribution for each individual is the same in each generation, but the
distribution can be varied from generation to generation or even from individual to individual, to model,
for example, public health control measures. Compared to SEIR models, which model a finite population,
branching process models do not consider depletion of susceptible individuals; all outbreaks either self-
extinguish or grow indefinitely. As a result, branching process models are only appropriate for the early
stages of an epidemic, when the number of infected individuals is negligible compared to the overall
susceptible population.
14.2.1 Probability Distribution Used in the Model
This study used an offspring distribution described by a negative binomial distribution with parameters R0
and k. Negative binomial distributions have been widely studied as models for nascent outbreaks and
have been used studying a variety of diseases, including influenza and SARS.1183,1184 In the distribution,
R0 is the commonly understood mean number of new cases each infected individual generates, and k
models the variation between infected individuals. The variable k incorporates differences between
individuals caused both by variations in social behavior as well as biological variation in, e.g., shedding
or infectious period. At low k values (< 0.5), individual variation is greater, and single individuals have
both a greater probability of generating many offspring, as well as a greater probability of generating
1183 Fraser C et al (2011) Influenza transmission in households during the 1918 pandemic. American journal of epidemiology
174: 505-514
1184 Lloyd-Smith JO et al (2005) Superspreading and the effect of individual variation on disease emergence. Nature 438: 355-
359

none. An outbreak described by a small value of k trends toward many outbreaks that self-extinguish
rapidly, with a small number of outbreaks that grow rapidly due to one or two individuals generating
many offspring. At larger k values, individual variation decreases, with each individual more likely to
generate a number of offspring near R0. A detailed description of the R0 and k values chosen for
influenza, SARS-CoV, and MERS-CoV is provided in the Supporting Information.
14.2.2 Construction of a Two-Type Model for Workers and Community Members
We modeled laboratory workers and community members separately within the BPM. In an outbreak
caused by a laboratory loss of control event, laboratory workers and community members are, on average,
likely to infect different numbers of individuals due to individual behavior choices, such as self-isolation,
as well as potentially different public health control measures and timings of control measures. For
example, lab workers are likely to recognize a disease is novel and laboratory-acquired, and may be more
likely to self-isolate when infected. Alternatively, if an outbreak is spreading between laboratory workers,
the institution or principal investigator may choose to shut down the lab and order isolation of all lab
workers, even those not yet showing symptoms, in order to stop the spread. Such drastic steps are
unlikely to be possible for entire communities, and community members are less likely to make major
changes to daily routines due to infection.
To incorporate these differences and track lab workers and community members separately, we used a
two-type branching process model. Two-type branching process models operate similarly to the standard,
one-type models, with the modification that the number of offspring distributions is expanded from one to
four, for each combination of parent type and offspring type. For a given overall outbreak described by R0
and k, we modified the single offspring distribution using values described in Table 14.2
Table 14.2. Modification of Offspring Distribution for Two-Type Model

Offspring Type
Lab Worker Community Member
Parent Type R0 value k value R0 value k value
Lab Worker w*R0 k/2 (1-w)*R0 k/2
Community Member 0 k R0 k
It was reasoned that a laboratory worker infected by a community member is likely to treat the infection
like any regular seasonal or community-acquired disease and not take any special steps to avoid spread,
and, additionally, if a laboratory loss-of-containment event has caused an outbreak significant enough that
community to community secondary spread is occurring, workers may no longer identify the outbreak as
novel or laboratory caused. For this reason, lab workers infected by community members are treated as
community members, and the R0 of community members infected laboratory workers is therefore fixed at
zero. As a result, from the perspective of a community member, the branching process model is largely a
one-type model, and the offspring distribution for community members is thus properly described by the
input R0 and k values.
For laboratory workers generating both types of offspring, two different probability distributions are used
for each type of offspring. The sum of these two distributions (i.e. the probability of generating a total
number of offspring) should be equivalent to a negative binomial probability distribution with parameters
R0 and k. The infinite divisibility theorem for negative binomial distribution states that a negative
binomial distribution, O, described by parameters R0 and k, can be broken into n independent negative
binomial distributions with parameters R0/n and k/n. For two distributions, W and C would each be

negative binomial distributions described by R0/2,k/2, O = W+C, where O describes the distribution
describing the total number of offspring, W describes the offspring distribution for workers generating
workers, and C describes the offspring distribution for workers generating community members.
Distributions W and C, as described above, have identical parameters, and thus workers would be equally
likely to generate workers and community members. However, workers are more likely to contact their
colleagues in the workplace than they are community members, and thus are more likely to infect
additional workers than community members. Based on a survey of individual contact frequencies
containing data on the location of the contact,1185 the fraction of contacts that laboratory workers would
have in the workplace versus elsewhere was estimated, and incorporated this into a parameter, w, where w
is the fraction of contacts that occur at work and (1-w) is the fraction that occur elsewhere. The R0 values
of the corresponding two-type offspring distributions are then multiplied by these factors. For the infinite
divisibility theorem to hold, the R0 values of each of the two-type offspring distributions must be equal,
and each half of the one-type distribution. However, for values of w near 0.5, the error between the one-
type distribution and sum of the two-type distribution (W+C-O) is small, where low numbers of offspring
are slightly less likely and large numbers of offspring slightly more likely in the two-type model, with a
maximum error of any specific offspring number of approximately 0.5%. On average, the total number of
offspring generated is slightly less for the sum of the two-type distributions than the one type distribution
but the difference is typically <0.5% of the total number of offspring generated, though statistical
fluctuations can result in more total offspring being generated by the sum of two-type distributions.
Figure 15.2 illustrates this difference for one million draws from each distribution, with R0= 1.3, k=1.0,
values appropriate for seasonal flu, and w=0.6951, the value used in all of our simulations.
1185 Mossong J et al (2008) Social contacts and mixing patterns relevant to the spread of infectious diseases. PLoS medicine 5:
e74

Figure 14.2. Difference in probability of generating different offspring. One million draws of number of
offspring were done for the one-type and each two-type distribution. The y-axis shows the difference between
the distributions as a fraction of the total number of draws.
14.2.3 Incorporation of Control Measures
Two types of control measures were incorporated into the model: population-wide control (i.e. social
distancing) and individual control (quarantine and isolation), based on previous work by Lloyd-Smith and
colleagues.1186 In the model used in this project, these types of control measures can be active in one of
five possible combinations as summarized by Table 14.3. We did not consider population-wide control on
laboratory workers in the absence of control on community members, because social distancing on just a
small section of the population is unprecedented and unwarranted. Additionally, individual control on the
community in the absence of control on laboratory workers was not considered, as it did not seem
reasonable to quarantine some individuals while avoiding quarantine on those most likely to be infected.
1186 Lloyd-Smith JO et al (2005) Superspreading and the effect of individual variation on disease emergence. Nature 438: 355-
359

Table 14.3. Control Measure Combinations
Number Control on Laboratory Workers Control on Community Members
1 Individual None
2 Individual Individual
3 Population Population
4 Individual & Population Population
5 Individual & Population Individual & Population
Each control measure, is implemented by modifying the corresponding offspring distribution via two
parameters: c, representing the strength of control, which vary from 0 (no control) to 1 (perfect or
absolute control), and g, the generation at which control becomes active, with the modifications to the
offspring distribution only present if the current generation is greater than or equal to g. In our model, the
control strengths of the two types of control may vary independently of each other, but the strength of a
particular control measure is the same for lab workers and community members. This reasoning was
based on the idea that public health and other responders to a nascent outbreak would be unlikely to, for
example, loosely quarantine lab workers and tightly control community members, or vice versa. (Recall
that the early stages of the response to loss-of-containment events on laboratory workers takes place
within the worker response event tree prior to the initiation of the BPM) The modifications to the
offspring distribution for each type of control are summarized in Table 14.4.
In population wide control, each individual reduces the number of contacts they have by a fraction given
by the control strength cp, resulting in a decrease in the mean number of people a person infects, and thus,
R0, by a factor (1-cp). However, the variation between individual’s infectiousness does not change,
resulting in the same k value prior to control being active.
In the type of individual control modeled here, a fraction, ci, of those infected individuals that would have
otherwise generated a non-zero number of offspring are isolated and thus instead generate none. This
measure has the net effect of increasing the proportion of zeros in the resulting offspring distribution. As
discussed in work by Lloyd-Smith and colleagues,1187 modeling of this control measure can be
accomplished in one of two ways: by directly modifying the resultant draws from the negative binomial
distribution and, with probability ci, setting the number of offspring for draws greater than zero to zero, or
by finding a solution to an alternative analytical equation, the solution to which gives a an approximate k
value assuming a negative binomial distribution, resulting in a negative binomial distribution with
parameters (1-ci)*R0 and ki that closely resembles the exact distribution under control for almost all of k
space.1188 In the former approach, the control measure is modeled exactly, but the effective value of k
used is not-knowable a priori. In the latter, the control measure is modeled with some error, but the exact
value of effective k is known, as it is specified in the negative binomial draws done under control. In the
approach used here, the latter approach was taken because computing the probability an outbreak self-
extinguishes requires knowing a value for k.
1187 Ibid.
1188 Ibid.

Table 14.4. Modification to the Offspring Distribution for Control Measure Types
Offspring Type
Lab Worker Community Member
Parent Type Control R0 Value k value R0 Value k value
Measure
Individual (1-ci)*w*R0 ki/2 (1-ci)*(1-w)*R0 ki/2
Lab Worker Population (1-cp)*w*R0 k/2 (1-cp) *(1-w)*R0 k/2
Both (1-ci)*(1-cp)*w*R0 ki/2 (1-ci)*(1-cp)*(1-w)*R0 ki/2
Individual 0 ki* (1-ci)*R0 ki
Community
Population 0 k (1-cp) *R0 k
Member
Both 0 ki (1-ci) *(1-cp) *R0 ki
*As R0 for community members generating workers is 0, the value of k has no effect on the resulting distribution
14.2.4 Terminating Models Due to Loss of Control
As mentioned in the main text, the BPM was used to model the initial stages of an outbreak, when it is
still circulating in the local community and has the potential to self-extinguish or be brought under
control. The simulations of each outbreak were terminated when one of four conditions was met:
• The outbreak self-extinguished (no new cases were generated by any infected individuals in the
current generation).
• Beginning in generations after all control types, if any, had been activated, the model calculated
that, given the number of cases in the current generation, that the outbreak had less than a 5%
chance of extinguishing at any point in the future.
• That any generation included 1,000 or more infected individuals.
• The outbreak had persisted for 200 generations without any of the above conditions being met.
The outbreak was considered out-of-control when any of the conditions other than self-extinguishing was
met. An outbreak that had less than 5% chance of self-extinguishing even after all control measures were
implemented was highly likely to grow to a size beyond that which local health officials could contain.
Even if an outbreak had a significant chance of self-extinguishing, 1000 simultaneous cases would likely
overwhelm the capacity of local resources to contain, and outbreaks would likely seed elsewhere prior to
when the outbreak self-extinguished. Finally, outbreaks were terminated after 200 generations to avoid
wasting a significant fraction of computational resources on simulations that, by stochastic chance, persist
for considerable lengths of time without meeting any other termination condition. This condition was
never reached in any simulation for 97.7% of the more than five million parameter combinations tested
and reached less than 5% of the time in 99.5% of parameter combinations tested, thereby having an
insignificant effect on the results. Because the outbreak had still not self-extinguished, these outbreaks
were considered out-of-control in order to conservatively estimate risk. However, should an outbreak
persist for 200 generations within a local community, travel of individuals in and out of the community
would result in a high likelihood of the outbreak seeding elsewhere, representing a local loss-of-control
event.

14.2.5 Calculation of Self-Extinguishing Probability
In a two-type branching process model with 2x2 matrices of R0 and k values, where R0,ij and kij represent
the values for type i individuals generating type j offspring, the probabilities of the outbreak self-
extinguishing at some future generation given one infected individual of each type in the current generation,
[q1 q2], where q1 is the probability of an outbreak with one individual of type 1 and q2 the same for a type 2
individual, can be derived from the basic principles of branching process models,1189 assuming R0,ij and kij
are time-invariant, with [q1 q2] as solutions for of the following system of equations:
−𝑘11 −𝑘12
𝑅11 𝑅12
(1 + ( ∗ (1 − 𝑞1 ))) ∗ (1 + ( ∗ (1 − 𝑞2 ))) − 𝑞1 = 0
𝑘11 𝑘12
−𝑘21 −𝑘22
𝑅21 𝑅22
(1 + ( ∗ (1 − 𝑞1 ))) ∗ (1 + ( ∗ (1 − 𝑞2 ))) − 𝑞2 = 0
𝑘21 𝑘22
Given the values of q1 and q2, and infected numbers of individuals I1 and I2, the overall probability of the
outbreak self-extinguishing is given by:
𝑞𝑡𝑜𝑡 = 𝑞1 𝐼1 ∗ 𝑞2 𝐼2
As the equations solved presume time-invariance in R0 and k, the outbreak was not presumed out of
control until qtot ≤ 0.05 at a generation after all control measures, and therefore modifications to R0 and k
had taken place.
14.3 Methodological Details of the HHS-BARDA IIM
14.3.1 Computation of region-specific contact rate matrices
Each of the twelve global regions simulated by the HHS-BARDA Interactive Influenza Model (referred to
as the IIM) used a different contact matrix that incorporated demographic differences between regions,
including age distribution, household size distribution, and school class size. For the region representing
high income countries in Europe and Central Asia (ECA), we used primary data gathered in a contact
survey of several countries within that region1190 to calculate a 4x4x6 three-dimensional matrix of mean
contact frequencies, F, where fijk is the expected daily number of contacts a person in age bracket i and
living in household size k would make with people in age bracket j. Reported contacts where either the
age of the reporter or person contacted, or household size of the reporter were unknown or omitted were
removed from the data set. Persons were grouped into 4 age groups: 0-4 years old, 5-19, 20-64, and 65+,
representing young children and infants, school-aged children, adults, and the elderly, respectively, based
on the default age groups used by BARDA. Households were divided into size 1-5, and 6+, based on the
groupings in the primary data. For each individual reporting contacts, the sum of the number of contacts
that individual made with persons of each age group were tracked and summed with other individuals of
the same age and household size, to create a 4x4x6 matrix of total number of contacts by age and
household size. This matrix was divided by the number of reporters within each age group and household
size to get the mean contact frequency matrix F.
1189 For a review of branching process models see Harris TE (2002) The theory of branching processes: Courier Corporation.
1190 Mossong J et al (2008) Social contacts and mixing patterns relevant to the spread of infectious diseases. PLoS medicine 5:
e74

This 4x6x4 matrix was reduced to a 4x4 matrix C, where cij represents the daily frequency of contact of
an individual of age i with persons of age j using a population weighted average of the per-household size
contact rates, using the following equation:
6
𝑐𝑖𝑗 = ∑ 𝑓𝑖𝑗𝑘 ∗ ℎ𝑘
𝑘=1
where hk represents the fraction of people living in a household of size k.
For high income ECA, the matrix C, after balancing (see below), was used as the contact matrix in the
IIM simulations for that region. For the other regions, the same matrix F was combined with region-
specific hk values to generate a new matrix C. In addition, each other matrix was modified to account for
local differences in population distribution by age and class size compared to high income ECA. We
assumed that the age-specific contact rates, cij vary proportionally to the fraction of the population of age j
(i.e., the more individuals of a certain age composing a community, the larger the frequency any
individual would contact one of them), by the following equation:
𝑐𝑖𝑗 = 𝑏𝑖𝑗 ∗ 𝑎𝑗
where aj is the fraction of people in the region simulated of age j, and bij is a scalar multiplier that remains
fixed across all regions. Using the matrix C for high income ECA, presumed to already be corrected by
the scalars b, corrected c values for other countries were calculated using the following relationship (and
using North America, abbreviated NA, as an example):
𝑎𝑗,𝑁𝐴
𝑐𝑖𝑗,𝐸𝐶𝐴
⁄𝑐𝑖𝑗,𝑁𝐴 =
𝑏𝑖𝑗∗𝑎𝑗,𝐸𝐶𝐴
⁄𝑏
𝑖𝑗∗𝑎 𝑗,𝑁𝐴
⇒𝑐 𝑖𝑗,𝑁𝐴
= 𝑐𝑖𝑗,𝐸𝐶𝐴 ⁄𝑎 𝑗,𝐸𝐶𝐴
Similarly, the contact frequency of school age children contacting school age children was corrected
using a class-sized based multiplier. We assumed that the contact rate of children with children, c22,
varied by:
𝑐22 = 𝑑 ∗ 𝑠
where s is the average class size within a region and d is a scalar multiplier again fixed across all regions.
Using a similar relationship to that of the age-specific correction above, the correction for class size
becomes:
𝑠
𝑐22,𝑁𝐴 = 𝑐22,𝐸𝐶𝐴 𝑁𝐴⁄𝑠𝐸𝐶𝐴
Each of these computations of cij were presumed to be multiplicative and independent such that, using the
rate of children contacting children in North America as an example, the overall calculation of c22
becomes:
6 𝑎𝑗,𝑁𝐴 𝑠
𝑐22,𝑁𝐴 = (∑ 𝑓22,𝑘 ∗ ℎ𝑘,𝑁𝐴 ) ∗ ⁄𝑎𝑗,𝐸𝐶𝐴 ∗ 𝑁𝐴⁄𝑠𝐸𝐶𝐴
𝑘=1
In addition to the corrections made above, each matrix C was also balanced. Given that every contact
involves two individuals, the total number of contacts all people of age i make with age j must also equal
the number of contacts people of age j make with age i. This can be represented in the contact matrix by:

𝑐𝑖𝑗 ∗ 𝑎𝑖 = 𝑐𝑗𝑖 ∗ 𝑎𝑗
Because the primary data contained no information on how to correct for any apparent imbalances in
overall contact rates, the matrices were balanced by assuming the overall contact numbers were equal to
the mean contact numbers of people of age i with age j and j with i. To accomplish this with each matrix
C, each element of the matrix, cij, was multiplied by the fraction ai of the population of that age to result
in a new matrix of proportional contact numbers D. That matrix was added to its transpose DT and then
each element was divided by 2 to get a balanced matrix of mean proportional contact numbers. Finally,
each element of the matrix was divided by the fraction ai of the population to convert the balanced contact
number matrix into a balanced contact rate matrix CB, where each element of CB is given by the following
equation:
(𝑐𝑖𝑗 ∗ 𝑎𝑖 + 𝑐𝑗𝑖 ∗ 𝑎𝑗 )
𝑐𝐵,𝑖𝑗 = ⁄
(2𝑎𝑖 )
These balanced contact rate matrices were used as inputs into the IIM.
14.4 Additional Data on the Potential Proliferation of GoF Research
Table 14.5. Terms Used to Query PubMed and Web of Science Databases.
[enhanced or increased] and [morbidity or mortality or pathogenicity] and [influenza virus]
[increased or enhanced] and [virulence] and [influenza virus]
[increased or enhanced] and [tropism] and [influenza]
[increased] and [human or mammalian] and [adaptation] and [influenza]
[immune system evasion] and [influenza]
[increased or enhanced] and [transmission] and [influenza virus]
[increased infectivity] and [influenza virus]
[h1n1] and [increased or enhanced] and [transmission or virulence or immune evasion or tropism or mortality or
morbidity or infectivity]
[h1n9] and [gain of function]
[h7n9] and [enhanced or increased] and [transmissibility or tropism or mortality or morbidity or viral production
or resistance or immune evasion]
[sars or severe acute respiratory syndrome]
[enhanced or increased] and [morbidity or mortality or pathogenicity] and [sars or mers]
[enhanced or increased] and [virulence] and [sars or mers]

Table 14.5. Terms Used to Query PubMed and Web of Science Databases.
[enhanced or increased] and [tropism] and [sars or mers]
[enhanced or increased] and [human or mammalian adaptation] and [sars or mers]
[enhanced or increased] and [immune system evasion] and [sars or mers]
[enhanced or increased] and [transmission] and [sars or mers]
[enhanced or increased] and [infectivity] and [sars or mers]
Figure 14.3: Authorship Relationships for Flu-1918 Case Study.1191
1191 Each dot represents a paper with an indicated last author. If an earlier middle author became a last author on a subsequent
paper with a different last author, a line was drawn between the dots.

Figure 14.4: Authorship Relationships for SARS-AM Case Study.

Figure 14.5: Authorship Relationships for Flu-PB2 Case Study.

15 Appendix IV. Benefit Assessment
Chapter 15 provides fully referenced, in-depth discussions of the potential benefits of GoF research
involving coronaviruses and influenza viruses. An overview of these benefits is provided in chapter 9.3
through 9.11 and 9.14.
15.1 Coronaviruses: Detailed Analysis of the Benefits of GoF Research 520

15.1.2 Overview of the potential benefits of GoF experiments involving coronaviruses 522
15.1.3 Benefits of GoF to Scientific Knowledge 524
15.1.4 Benefits of GoF-derived Model Systems 538
15.1.5 Benefits of GoF to Public Health/Medicine 548
15.2 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances Virus
Production 565
15.2.1 Overview of the GoF Landscape: Approaches that Enhance the Production of Influenza
Viruses 565
15.2.2 Overview of the potential benefits of GoF approaches that enhance the production of
influenza viruses 566
15.2.3 Benefits of GoF Research that Enhances Production of Influenza Viruses to Current Vaccine
Production Practices 568
15.2.4 Benefits of GoF Research that Enhances Production of Influenza Viruses to Scientific
Knowledge and to Future Influenza Vaccine Production Practices 579
15.3 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Enhances Mammalian
Adaptation and Transmissibility 596
15.3.1 Overview of Influenza GoF Landscape 596
15.3.2 Overview of the potential benefits of GoF approaches that enhance the fitness/transmissibility
of influenza viruses 597
15.3.4 Benefits to Surveillance 624
15.3.5 Benefits to decision-making in public health policy 639
15.4 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances Virulence 651
15.4.2 Overview of the potential benefits of GoF experiments involving coronaviruses 652
15.4.4 Benefits of GoF to Surveillance 685
15.4.5 Benefits of GoF to the Development of Vaccines and Therapeutics 685
15.4.6 Benefits to decision-making in public health policy 692
15.5 Influenza viruses: Detailed analysis of the benefits of GoF research that leads to evasion of existing
natural or induced adaptive immunity 693
15.5.1 Overview of the Influenza GoF Landscape 693
15.5.2 Overview of the potential benefits of GoF experiments that may lead to the generation of
influenza viruses that evade existing natural or induced adaptive immunity 694
15.5.4 GoF Benefits to Surveillance 707
15.5.5 GoF Benefits to the production of vaccines 715

15.6 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Leads to Evasion of
Vaccines 728
15.6.2 Overview of the Potential Benefits of GoF Experiments that may Lead to the Generation of
Influenza Viruses that are Resistant to Therapeutics 728
15.6.3 Benefits to Vaccine Development 729
therapeutics 730
Influenza Viruses that are Resistant to Therapeutics 731
15.7.3 Benefits to Scientific Knowledge 732
15.7.4 GoF Benefits to Surveillance 746
15.7.5 Benefits to Decision-Making in Public Health Policy 751
15.7.6 GoF Benefits to the Development of vaccines 753
15.7.7 GoF Benefits to the Development of Therapeutics 754
15.8 Influenza viruses: Detailed analysis of the benefits of GoF research involving reassortment 760
15.8.2 Overview of the potential benefits of GoF experiments involving reassortment 761
15.8.4 Benefits of GoF approaches to Surveillance 773
15.8.5 Benefits to decision-making in public health practice and policy 775

15.9.1 Summary of Findings 777
15.9.3 Potential Benefit 1- Improvements in the design and production of vaccines 781
15.9.4 Potential Benefit 2- Assistance in the development of new influenza or coronavirus
antivirals 796
15.9.5 Potential Benefit 3- Benefits to pandemic preparedness planning 807
15.9.6 Information on influenza vaccine production in low- and middle-income countries 815
15.10 List of Subject Matter Experts Interviewed for the Benefit Assessment 821

15.1 Coronaviruses: Detailed Analysis of the Benefits of GoF Research
15.1.1 Introduction
15.1.1.1 Scope of assessment
This assessment describes the benefits of GoF experiments involving SARS-CoV, MERS-CoV, and
SARS/MERS-like bat CoVs. From a review of the coronavirus literature, experimental approaches that
are reasonably anticipated to lead to the following phenotypic changes were identified:

• Altered host range (typically accompanied by enhanced virulence in the new host),
• Enhanced fitness or virulence in cell culture or laboratory animal model systems, respectively and
natural or induced immunity. Finally, no coronavirus research that is reasonably anticipated to lead to
evasion of diagnostics or of vaccines in development was identified. (It should be noted that there are
currently no FDA-approved vaccines or therapeutics for coronaviruses.)
The four human coronaviruses that cause mild to moderate respiratory illnesses such as the common cold
or croup (coronaviruses HKU1, OC43, 229E, and NL63) were not evaluated because these are not
considered in the NSABB GoF Framework. Throughout this report, use of the term “coronaviruses” or
“CoVs” refers specifically to SARS-CoV, MERS-CoV, and SARS/MERS-like bat CoVs such as HKU4
and HKU5.
15.1.1.2 Overview of Coronavirus GoF Landscape
Here, a brief overview of the experimental approaches within each GoF phenotypic category is provided.
Each approach will be discussed in more detail in the context of detailed analysis of the benefits of GoF
research involving coronaviruses, below.
15.1.1.2.1 Experimental approaches that lead to enhanced pathogen production
Serial passaging of CoV in cell culture leads to the generation of higher-yield viruses. This approach is
used to enhance the growth of viruses with naturally poor growth properties, in order to develop an in
vitro model system for experimental use.
15.1.1.2.2 Experimental approaches that alter host tropism in mammals

A second approach that leads to altered host range involves serial passaging of CoVs in mice, which leads
to the generation of viruses that have adapted to more efficiently infect and cause disease in mice. The
purpose of this experiment is two-fold:
A final approach involves targeted mutagenesis to introduce mutations that are associated with altered
host tropism, which is performed to demonstrate that the mutation(s) are necessary and sufficient to alter
host tropism. As above, this information provides a foundation for follow-up studies investigating the
phenotypic traits underlying virus adaptation to new hosts.
15.1.1.2.3 Experimental approaches that enhance fitness or virulence in cell culture or laboratory animal
model systems

and sufficient to enhance virulence. As above, this information provides a foundation for follow-up
studies to elucidate the mechanistic basis of virulence.
A third approach involves serial passaging of attenuated viruses that are candidate live attenuated
vaccines (LAVs), in order to determine whether the viruses acquire mutations that enhance
fitness/virulence. Because LAVs with an ability to recover fitness during growth in vivo could cause
adverse outcomes in people, a negative result is an important indicator of safety for any live attenuated
vaccine in development.
15.1.1.2.4 Experimental approaches that lead to evasion of therapeutics in development
that are resistant to inhibition/neutralization by that therapeutic. The purpose of the experiment is to
understand whether and how readily resistance will arise in response to selective pressure from the
therapeutic and to identify mutations that are associated with resistance to the therapeutic, which provides
a foundation for follow-up studies investigating the mechanisms underlying antiviral activity and antiviral
resistance. This information benefits the development of these therapeutics. Specifically, emergence-of-
resistance data speak to the potential field efficacy of the therapeutic, and information on both antiviral
mechanism and emergence of resistance are important components of an investigational new drug
application to the FDA.
15.1.2 Overview of the potential benefits of GoF experiments involving coronaviruses
This section evaluates whether any of the GoF CoV approaches have the potential to benefit each of the
general benefit areas described in the NSABB’s “Framework for Conducting Risk and Benefit
Assessments of Gain-of-Function Research.” Also described are additional benefit areas identified during
research. Each potential benefit will be analyzed in detail below.
GoF approaches have the potential to directly benefit scientific knowledge by providing insight into the
mechanisms underlying adaptation of coronaviruses to new hosts as well as the mechanistic basis of
coronavirus virulence. In addition, the development of animal models using GoF approaches has the
potential to indirectly benefit scientific knowledge by enabling the study of disease pathogenesis,
including the role of host factors in disease pathology.
Currently, GoF approaches do not have the potential to benefit public health, agricultural animal, or
wildlife surveillance. Although CoV researchers stated that they could envision using information about
the molecular determinants of human adaptation and virulence to assess the risk posed by animal CoVs
circulating in nature, similar to the influenza field, this application is currently unfeasible for two reasons:
(1) CoV surveillance networks are extremely limited, with large gaps in coverage in humans and animals,
and (2) the state of knowledge about the molecular determinants of human adaptation and virulence is
poor.1192
1192
For example, out of more than 1700 bat species, only ten have been surveilled for evidence of CoV infection
(and those ten on an ad hoc rather than a systematic basis).

15.1.2.3 Vaccines
GoF approaches have the potential to benefit the development of vaccines in three ways:
• GoF approaches that lead to the discovery of virulence factors identify potential gene targets for
attenuation, for the development of live attenuated vaccines (LAVs),
• Serial passaging of LAV strains in animals is used to test whether strains recover virulence upon
growth in vivo, which is an important aspect of vaccine safety, and
• GoF approaches that lead to the development of animal-adapted viruses (i.e. serial passaging of
viruses in laboratory animals to alter host tropism and enhance virulence) enable the testing of
vaccine candidates in animal models that mimic the pathology of human disease.
GoF approaches have the potential to directly benefit the development of therapeutics in several ways:
• GoF approaches that lead to the discovery of virulence factors identify potential new therapeutic
targets,
• GoF approaches that lead to evasion of therapeutics in development provide information about
the potential field efficacy of the therapeutic and the mechanism of activity of the therapeutic,
both of which are critical components of an Investigational New Drug application to the FDA,
• GoF approaches that lead to evasion of therapeutics in development can provide insight into the
therapeutic dosing regimens and combination therapies (e.g. cocktails of monoclonal antibodies)
that are the least likely to permit evolution of resistance, and
• GoF approaches that lead to the development of animal-adapted viruses enable the testing of
therapeutic candidates in animal models that mimic the pathology of human disease.
15.1.2.5 Diagnostics
As diagnostic targets for CoVs are well-established, no potential benefits of GoF approaches to the
development of diagnostics were identified.1193,1194,1195,1196
1193
The FDA-approved diagnostic test for MERS-CoV targets two regions in the CoV genome: a region upstream
of the E gene (upE) and the reading frame 1a (orf1a). SARS can be detected through RT-PCR with sequences in
the polymerase 1 B region (pol 1B) and an adjacent downstream region of the genome as the targets. Other
diagnostic tests target sequences in the nucleocapsid (N) gene.
1194
Stephen M. Ostroff Acting Commissioner of Food and Drugs. Letter of Authorization RealStar® MERS-CoV
RT-PCR Kit U.S. .
http://www.fda.gov/downloads/MedicalDevices/Safety/EmergencySituations/UCM455348.pdf. Last Update
July 17, 2015. Accessed December 2015.
1195
Richardson SE et al (2004) The laboratory diagnosis of severe acute respiratory syndrome: emerging laboratory
tests for an emerging pathogen. The Clinical biochemist Reviews / Australian Association of Clinical
Biochemists 25: 133-141
1196
Mahony JB et al (2004) Performance and Cost evaluation of one commercial and six in-house conventional and
real-time reverse transcription-pcr assays for detection of severe acute respiratory syndrome coronavirus. J Clin
Microbiol 42: 1471-1476

15.1.2.6 Informing Policy Decisions
Because the US government is not actively engaged in public health preparedness activities for CoV
outbreaks and because there are no FDA-approved vaccines or therapeutics for CoVs, GoF approaches do
not have the potential to benefit decision-making in public health policy (e.g. informing countermeasure
stockpiling decisions, guiding decisions about strain selection for vaccine development, etc.)
GoF benefits to the development of vaccines and therapeutics could have downstream economic benefits.
Economic benefits were not explicitly evaluated in this report.
Below the potential benefits in all the fields identified above are analyzed: scientific knowledge, vaccines,
and therapeutics. For each field, the potential benefits of GoF approaches as well as the potential benefits
of alternative experimental approaches and alternative scientific and technical innovations that can
provide the same or similar benefits are analyzed. For each potential benefit, the scientific, technical, and
regulatory barriers to the realization of that benefit were identified; these impact the likelihood and timing
of the realization of the benefit. Next, the potential benefits of GoF research relative to alternative
approaches are evaluated, considering the barriers to the realization of the benefits of each.
This analysis is split into three sections. First, the potential for GoF approaches to directly benefit
scientific knowledge, including knowledge about mechanisms underlying the cross-species adaptation
and pathogenesis of coronaviruses, is evaluated. In this section, alternative experimental approaches that
can provide the same or similar information as GoF approaches are considered. Second, the potential
benefits of using model systems developed using GoF approaches are analyzed; these include benefits to
basic science research as well as to medical countermeasure (MCM) development. In this section,
alternative model systems that do not involve GoF approaches are evaluated (e.g. use of a naturally
susceptible host in lieu of using a virus adapted to a laboratory animal). Finally, the potential for GoF
approaches to directly benefit public health is assessed; this includes benefits to the development of
vaccines and therapeutics. In this section, alternative experimental approaches as well as alternative
scientific and technical innovations that have the potential to similarly benefit MCM development are
evaluated.
15.1.3 Benefits of GoF to Scientific Knowledge
Several GoF approaches generate information that directly benefits scientific knowledge by providing
insight into critical unanswered questions about coronavirus biology. Specifically, GoF approaches that
alter host tropism can provide insight into the mechanistic basis of cross-species adaptation, and GoF
approaches that enhance virulence in animal models enable the identification of virulence factors and
deepen understanding of the mechanisms underlying pathogenicity. In this section, the benefit of GoF
approaches to each of these scientific areas, relative to alternative experimental approaches that can
provide the same or similar scientific information, are discussed.
15.1.3.1 Scientific Knowledge Gap 1: How do animal coronaviruses adapt to humans? What are the
genetic and phenotypic traits underlying adaptation to humans?
SARS and MERS unexpectedly emerged from their animal reservoirs to infect humans in 2002 and 2012,
respectively. Surveillance of bats and other CoV reservoir species indicates that there is a large diversity
of animal CoVs circulating in nature, including many species that are genetically related to SARS and

MERS and thus may have the potential to spill over into human populations in the future.1197,1198,1199,1200
Although multiple coronaviruses have been shown to exhibit a flexible capacity for cross-species
transmission,1201,1202 the mechanisms underlying CoV adaptation to new host species are poorly
understood. Specifically, large gaps in knowledge remain regarding:
• The mechanistic basis of cross-species adaptation – what viral factors are involved, and what
phenotypic changes must occur in order for a CoV to adapt to efficiently infect and cause disease
in a new host species?
• The evolutionary mechanisms driving cross-species adaptation – what selective pressures drive
adaptation to new host species, and what is the order of acquisition of new genetic/phenotypic
traits needed for adaptation? And
• Whether the ability to adapt to new species is a conserved feature of all CoVs, and if so, whether
the mechanisms underlying adaptation of different CoV species are similar or distinct?
15.1.3.1.1 Potential Benefits and Limitations of GoF Approaches
Several GoF approaches can provide insight into these questions. Serial passaging of CoVs in cells
derived from a non-natural host organism or in a non-natural laboratory animal host selects for viruses
that more efficiently infect cells/animals, thereby enabling the identification of mutations that are
sufficient for adaptation to a new host species. Currently in the CoV field, these experiments involve
passaging of animal or zoonotic CoVs (such as MERS-CoV) in human cells or passaging of MERS-CoV
in mice. (SARS-CoV was previously adapted for growth in mice through serial passaging.) Identifying
where mutations arise during adaptation to new hosts points to viral factors that may play a role in
adaptation, and studying the phenotypic consequences of the mutations provides insight into the
mechanistic basis of cross-species adaptation. Of note, serial passaging in simple, in vitro model systems
provides more limited information about mechanisms underlying cross-species adaptation than serial
passaging in animals, and the phenotypic changes needed to adapt viruses for growth in cell culture may
not be relevant for in vivo adaptation. Analyzing viral sequences at multiple stages of in vivo passaging
can provide insight into the order of acquisition of genetic changes as well as information about mutations
that are positively and negatively selected over the course of adaptation. One key benefit of this approach
is that it can lead to the discovery of novel genetic traits and virus proteins that are involved in the process
of adapting to new hosts without the need for prior knowledge of viral adaptation factors. Moreover, this
approach can be used to explore the adaptation of any virus to a new host species, provided that the virus
can be grown in an appropriate model system. Finally, repeating the serial passaging experiment multiple
times with the same starting virus can provide insight into the mutational landscape of cross-species
adaptation – that is, whether the same changes tend to occur or whether there are multiple evolutionary
pathways for adapting to a new host species. The main limitations of this approach are that traits that
promote growth in a particular cell type or a non-human mammal may not be required for enhancing the
1199 Pfefferle S et al (2009) Distant relatives of severe acute respiratory syndrome coronavirus and close relatives of human
coronavirus 229E in bats, Ghana. Emerging infectious diseases 15: 1377-1384
535-538
1201 Baric RS et al (1999) Persistent infection promotes cross-species transmissibility of mouse hepatitis virus. Journal of
virology 73: 638-649
1202 Chen W et al (2005) SARS-associated coronavirus transmitted from human to pig. Emerging infectious diseases 11: 446-
448

ability of the virus to infect and transmit between humans and that laboratory evolution may not mimic
natural selection. Additionally, serial passaging identifies traits that are sufficient but may not be
necessary for adaptation to new hosts, and results gleaned from the one or two strains under study may
not be conserved in other CoV species.
Another GoF method for studying cross-species adaptation involves “Spike swapping” – that is, targeted
genetic modification to replace all or part of the CoV Spike protein, a surface protein that mediates virus
entry into cells and is a critical determinant of host restriction, with the Spike protein from another CoV
species. These experiments are considered gain-of-function because they are expected to alter host
tropism in mammalian species. The purpose of these experiments is two-fold. First, testing the ability of
chimeric CoVs to infect various types of cells and animals reveals the breadth of host tropism conferred
by a given Spike protein, and comparing the sequences of parental and donated Spike proteins with
different host tropism can uncover amino acid residues that mediate host restriction. Second, defining the
host tropism of animal CoVs and the number of amino acid changes that are needed to confer the ability
to infect human cells provides insight into whether the ability to adapt to new species is a conserved
feature of CoVs, as well as which animal CoVs are poised to spill over into human populations. (Of note,
these high-risk bat CoVs can then be targeted as part of efforts to develop broad-spectrum vaccines and
therapeutics, which will be discussed further in section 16.1.4.) The main drawback of this approach is
that it is limited to studying the role of the Spike-receptor interaction, and no other viral factors, in host
tropism. Another drawback is that chimeric “SARS plus animal CoV Spike” viruses may behave
differently from wild type animal CoVs; however, presenting an animal CoV Spike in the context of the
SARS virus better mimics the wild type virus than pseudotyping systems using other viruses, an
alternative approach discussed below. (Pseudotyping is the process of expressing the envelope protein or
surface glycoprotein from one virus on the surface of a different virus, e.g. replacement of the vesicular
stomatitis virus glycoprotein (VSV G) with the CoV Spike, enabling expression of the CoV Spike on the
surface of VSV. Pseudotyping is performed to study the function of the foreign virus protein in isolation,
as a risk mitigation measure, and/or to study the activity of a protein from a virus that is difficult to
culture, such as bat CoVs.)
Finally, targeted genetic modification of wild type viruses to introduce mutations that are associated with
adaptation to new hosts demonstrates that such markers are necessary and sufficient to broaden or alter
host tropism. Of note, these mutations can be discovered through a GoF approach, such as serial
passaging, or an alt-GoF approach, such as comparative sequence analysis (discussed below). This
information provides a strong foundation for follow-up studies investigating the mechanistic basis of the
adaptation phenotype.
15.1.3.1.2 Potential Benefits and Limitations of alt-GoF Approaches
Alternative experimental approaches can also be used to discover genetic traits associated with cross-
species adaptation of CoVs. First, comparing the sequences of CoVs with different species tropism,
including comparison of animal CoVs versus SARS/MERS and comparison of animal strains from
different geographic regions where spillover into human populations has and has not occurred (or has
occurred with different frequencies), can elucidate genetic traits that are associated with adaptation to
different hosts. Second, comparative sequence analysis of human CoVs from different time points during
an outbreak reveals how zoonotic CoVs adapt to humans following an initial spillover event. Relative to
the laboratory methods described above, this approach may be more likely to uncover conserved
determinants of cross-species adaptation because it involves analysis of multiple sequences, and analysis
of human isolates is more likely to identify traits that are relevant for adaptation to humans under natural
selective pressures. Importantly, follow-up studies are needed to confirm that the identified genetic traits
are responsible for altered host tropism.

Both types of comparative sequence approaches suffer from several significant limitations. First, the
success of comparative sequence analysis is constrained by the quality and availability of existing genetic
surveillance data. Relatively few sequences are available from relevant animal reservoirs, including bats
and camels (for MERS). The only published camel sequences are from the Middle East, precluding the
study of camel viruses from different geographic regions where spillover has/has not occurred. For the
study of human epidemic CoVs, a limited number of SARS sequences are available from the 2002 – 2003
outbreak, and because MERS transmission chains have been relatively short, MERS data are of limited
utility for studying adaptation mechanisms in humans. Of note, analysis of SARS epidemic strains reveals
only one evolutionary pathway for adaptation to humans, which may represent one of several possible
mechanisms. A second limitation is that, due to the large size of the CoV genome (27-32 kb) and the
genetic diversity of coronaviruses in nature, there are a very large number of genetic differences between
any two CoV strains, only a subset of which are likely to be important for cross-species adaptation.1203
Because of that “noise,” sequence comparisons are realistically limited to known regions of interest,
precluding discovery of novel factors that are involved in host adaptation. Due to the fact that only a few
proteins have been shown to be involved in cross-species adaptation and the function of most CoV
proteins is unknown, this limited focus represents a critical shortcoming of the comparative sequence
analysis approach. Although this limitation could be partially addressed by comparing sequences of
paired animal and human isolates (e.g. MERS isolates from infected humans and the camels that are the
likely sources of the infection), few such paired sequences are available. Third, this approach is reactive,
limited to the study of mechanisms underlying adaptation of CoVs that have already evolved to broaden
or alter their host tropism (e.g. SARS and MERS). The mechanisms driving adaptation of other CoVs to
new hosts may be different. Of note, MERS does not efficiently infect and transmit in humans, unlike
SARS, thus analysis of MERS sequences is limited to the discovery of traits that are associated with
partial adaptation to humans. Finally, analysis of historical sequences cannot identify traits that were lost
or negatively selected during adaptation (i.e. evolutionary pathways not taken) and thus provides a static
view of evolutionary mechanisms underlying cross-species adaptation.
Conceptually similar to “Spike swapping” experiments, several alternative approaches seek to define the
breadth of host tropism conferred by a given Spike protein. The first approach involves testing whether
MERS- or SARS-CoVs can infect cells derived from various non-human host species such as bats or cells
that do not naturally express CoV receptor proteins but have been engineered to ectopically express
receptor proteins from various species. This approach cannot be used for most animal CoVs, which
cannot be grown efficiently in cell culture to produce infectious material for laboratory assays.
Alternatively, two virus-free approaches can provide information about compatible Spike-host
interactions: (1) in vitro binding assays using recombinant Spike proteins and host receptor proteins from
different species, and (2) cell culture-based binding and virus entry assays using non-CoVs (e.g. murine
leukemia virus) that are pseudotyped with CoV Spike proteins. These in vitro systems can also be used to
confirm that amino acid substitutions in the Spike protein are necessary and sufficient to alter host
receptor binding and cell entry capabilities.
The major limitation associated with these virus-free approaches is that results may not be recapitulated in
the context of the wild type virus, as the virus context influences presentation of surface epitopes. CoV
researchers reported cases of both false positive and false negative results when using pseudotyped
viruses compared to wild type viruses.1204 Additionally, results from either virus-free approach may not be
conserved in a different strain context, and traits that promote binding of pseudotyped viruses to a
particular cell type may not be critical for adaptation to human hosts. Finally, these approaches are
1204 (2015b) Interviews with coronavirus researchers.

currently used to investigate the role of the Spike-receptor interaction in host restriction only and are
fundamentally limited to the investigation of known mammalian adaptation factors.
Structural modeling of Spike-receptor interactions, based on crystal structures of Spike-receptor

complexes, can also be used to identify amino acid residues in the Spike protein that may be important
determinants of host restriction. Though useful for generating hypotheses about mutations that may alter
host tropism, all predictions must be experimentally confirmed.
In principle, loss of function (LoF) approaches could also be used to study mechanisms underlying cross-
species adaptation, through the identification of genetic traits that are necessary for efficient infection of a
particular host (i.e. screening mutants for reduced infectivity). However, because SARS, MERS, and bat
CoVs do not naturally cause disease small laboratory animals, LoF approaches are not viable for the study
of mechanisms underlying cross-species adaptation using wild type viruses. Notably, LoF approaches
have been used to explore the genetic traits that are necessary for the mouse-adapted SARS strain to
efficiently infect mice, by reverting adaptive mutations individually and in combination using site-
directed mutagenesis and characterizing the infectivity of mutants.1205 However, the mouse-adapted strain
was originally generated using GoF approaches (i.e. serial passaging of SARS-CoV in mice). Although
cell culture systems could, in principle, be used for LoF studies involving SARS and MERS, in vitro
studies can provide minimal information about cross-species adaptation because the interaction of a virus
with the host immune system is a critical facet of adapting to new hosts. No LoF studies using cell culture
systems were identified in the scientific literature.
15.1.3.1.3 Summary – Benefits of GoF approaches relative to alt-GoF approaches
The scientific knowledge benefits and limitations of all GoF and alt-GoF approaches discussed in this
section, with respect to the ability of each approach to provide insight into the mechanisms underlying
cross-species adaptation, are summarized in Table 15.1. Together, this analysis reveals that serial
passaging, a GoF approach that alters host range, is uniquely capable of identifying novel viral genetic
traits and factors that contribute to cross-species adaptation. Moreover, to elucidate the molecular
mechanisms underlying the role of the Spike-receptor interaction in host adaptation, testing the
phenotypic consequences of mutations in animal CoV Spike proteins in the context of a chimeric virus
generated through GoF approaches provides a higher level of certainty in the validity of the results than
similar confirmatory experiments using recombinant proteins or pseudotyped viruses. However,
laboratory results in model systems may not translate to adaptation of viruses to humans in nature.
Conversely, sequence comparisons, an alt-GoF approach, are uniquely capable of identifying genetic
traits that are associated with mammalian adaptation across a variety of strains as well as discovering
genetic markers that are definitively associated with human adaptation. However, the causality of markers
identified through sequence analysis must be confirmed with a GoF experiment, and the utility of the
comparative sequence approach is severely compromised by the poor state of genetic surveillance for
CoVs in human and animal populations and the fact that it is limited to analysis of strains that have
caused human infections.
1205 Frieman M et al (2012) Molecular determinants of severe acute respiratory syndrome coronavirus pathogenesis and
virulence in young and aged mouse models of human disease. Journal of virology 86: 884-897

Table 15.1. Comparison of GoF approaches and corresponding alt-GoF approaches that benefit scientific knowledge: How do animal CoVs adapt to
humans? What are the genetic and phenotypic traits underlying adaptation to humans?
Experimental approach Benefits Limitations
• Associative – whether mutations are necessary
for adaptation must be experimentally confirmed
• Identify novel genetic traits that are • Simplicity of model system – provides limited
sufficient to alter host tropism, for any information about cross-species adaptation, and
GoF #1 [4a]: (In vitro approach) Serial passaging of virus results may not be relevant for in vivo adaptation
virus in cells derived from non-natural host organism • Identify novel viral factors that are • Translatability – results from model systems may
involved in cross-species adaptation, for not translate to human infections
any virus
• Narrow breadth – results may not generalize to
other CoV strains
• Identify novel genetic traits that are • Associative – whether mutations are necessary
sufficient to alter host tropism, for any for adaptation must be experimentally confirmed
virus • Translatability – results from model systems may
GoF #2 [4b]: (In vivo approach) Serial passaging of • Identify novel viral factors that are not translate to human infections
virus in non-natural host organism (e.g. mice) involved in cross-species adaptation, for • Artificiality – lab-directed evolution may not
any virus mimic natural selection
• Provides in-depth information about • Narrow breadth – results may not generalize to
the evolutionary mechanisms other CoV strains
underlying cross-species adaptation
GoF #3 [5,6]: Targeted genetic modification to replace
all or part of the CoV Spike protein with the Spike
protein from another CoV species • Define the breadth of host tropism • Limited to studying the role of the Spike protein
conferred by a particular Spike protein in cross-species adaptation
• Animal CoV + SARS Spike
• Identify amino acid substitutions within • Chimeric viruses may behave differently than
• SARS/MERS CoV + animal CoV Spike
the Spike protein that may mediate host wild type viruses
restriction • Associative – whether substitutions are necessary
Characterize phenotypic properties of chimeric virus and and sufficient for host restriction must be
• Gain insight into the potential for bat
compare sequences of animal CoV and SARS/MERS experimentally confirmed
CoVs to adapt to humans
Spike proteins

• Identify genetic traits that are necessary
and sufficient to alter host tropism
GoF #4 [7]: Targeted genetic modification of virus to
• Confirm viral factors that are involved • Translatability – results from model systems may
introduce mutation(s) shown to be associated with
in cross-species adaptation not translate to human infections
adaptation to new hosts
• Gain insight into mechanisms • Narrow breadth – results may not generalize to
• Characterize ability of mutant virus to infect new cell underlying virus adaptation to new other CoV strains
type or animal
hosts, including identification of
underlying phenotypes
• Utility and success of approach is constrained by
the quality and availability of genetic surveillance
data
Alt-GoF #1 [4]: (virus free) Comparative analysis of
• Identify genetic traits that are associated • Bias – limited to investigation of known genetic
surveillance data to identify genetic markers associated
with adaptation to humans under regions of interest
with adaptation to humans:
natural selective pressures • Reactive – limited to the study of CoVs that have
• Animal versus human epidemic strains o Identify conserved traits, if large already caused human infections (i.e. SARS and
• Animal strains from different geographic regions numbers of sequences are analyzed MERS)
where spillover of animal virus into human population • Gain insight into the evolutionary • Associative – whether mutations are necessary
has/has not occurred mechanisms underlying cross-species and sufficient for adaptation must be
• Human epidemic CoV strains from different adaptation experimentally confirmed
timepoints during an outbreak • Static – evolutionary insight is limited because
historical isolates represent discrete events along
an evolutionary continuum

Alt-GoF #2 [6]: Test whether SARS or MERS can infect
cells derived from non-human host species (e.g. bat, • Define the breadth of host tropism
mouse, etc.) or can infect receptor-null human cells that conferred by a particular Spike protein • Approach cannot be used for animal CoVs that
are ectopically expressing receptor proteins • Identify amino acid substitutions within cannot be grown efficiently in cell culture
the Spike protein that may mediate host • Associative – whether substitutions are necessary
• Test whether cells can be infected with animal-origin
restriction and sufficient for host restriction must be
virus
• Compare sequences of human and animal-origin host experimentally confirmed
receptors to identify amino acids associated with host
restriction
Alt-GoF #3 [5,8]:
• Simplicity of model system – results may not be
• In vitro, virus free: recapitulated in the context of the wild type virus
o In vitro binding assays using recombinant CoV
• Define the breadth of host tropism
conferred by a particular Spike protein • Narrow breadth - results may not generalize to
Spike proteins and host cell receptor proteins other CoV strains
• In cells, pseudotyped virus: • Identify amino acid substitutions that
are necessary and sufficient to alter host • Translatability – traits that promote binding to a
o Test host cell binding and entry using virus particular cell type may not be critical for
pseudotyped with CoV Spike proteins tropism
adaptation to human hosts
• Targeted genetic modification of Spike proteins to • Limited to studying the role of the Spike protein
introduce mutations associated with altered host range in cross-species adaptation
in either context
• Predictive – phenotypic consequences of
substitutions must be experimentally confirmed
Alt-GoF #4 [7]: (in vitro, virus free) Structural modeling • Predict amino acid substitutions within • Simplicity of model system – may not reflect
of Spike-receptor interactions, based on crystal structures the Spike protein that may mediate host virus-host cell interactions
of Spike-receptor protein complexes restriction
• Limited to studying the role of the Spike protein
in cross-species adaptation
* GoF and alt-GoF approaches are listed in numerical order. Numbers in brackets specify experimental approaches described in the landscape tables
(Supplemental information).

15.1.3.2 Scientific knowledge gap 2: How do SARS and MERS coronaviruses cause disease? What are
the critical viral virulence factors and viral genetic determinants of virulence?
Why SARS and MERS coronaviruses cause severe respiratory infections while other human
coronaviruses cause mild to moderate illness is unknown.1206 Specifically, the viral genetic and
phenotypic traits underlying the enhanced pathogenicity of SARS and MERS relative to other human
coronaviruses are poorly understood, and only a few viral virulence factors have been identified and
characterized (such as the CoV Spike protein, which mediates viral entry into host cells). As there are no
FDA-licensed vaccines or therapeutics for SARS or MERS, research in this area is important not only for
increasing basic science knowledge about coronavirus biology but also for identifying potential new
targets for therapeutics or for attenuation, for the purpose of developing live attenuated vaccines (LAVs).
This benefit to MCM development will be discussed in more detail in section 17.1.4, below.
Serial passaging of CoVs in cell culture or laboratory animals, which selects for enhanced fitness (in
vitro) or enhanced virulence (in vivo), is a GoF approach that enables the identification of mutations
associated with enhanced fitness/virulence. Identification of virulence-associated mutations can lead to
the discovery of new viral virulence factors and provides a foundation for follow-up studies investigating
the mechanistic basis of the enhanced fitness/virulence phenotype observed in emergent viruses. As
above, a key benefit of this approach is the ability to generate and identify novel mutations and viral
proteins that contribute to fitness/virulence, without prior knowledge about viral virulence factors.
Moreover, this approach can be performed with any coronavirus that is capable of infecting appropriate
cell culture or animal model systems, including SARS-CoV, MERS-CoV, and chimeric animal-human
CoVs used as tools for the study of animal CoVs that cannot be grown in model systems (discussed
further below). In vivo serial passaging can provide a wider breadth of information than the in vitro
approach because replicative fitness, though a component of virulence, does not necessarily correlate with
virulence in vivo. For example, infected animals that are symptomatic and asymptomatic may exhibit
similar viral loads, demonstrating that disease pathology is not simply caused by viral replication but also
by the interaction of a virus with the host immune system. The roles of complex host-virus interactions
can only be studied in the context of an animal model system (although underlying phenotypes can be
studied in vitro). For both in vitro and in vivo model systems, insights may not translate to human
infections, and viral factors and phenotypes that contribute to virulence in the CoV strain under study may
not generalize to other CoV strains.
A second GoF approach for studying virulence involves targeted genetic modification of wild type viruses
to introduce mutations that are associated with enhanced fitness/virulence, which demonstrates that such
markers are necessary and sufficient to enhance fitness/virulence. Of note, these mutations can be
discovered through a GoF approach, such as serial passaging, or an alt-GoF approach, such as
comparative sequence analysis (discussed below). This information provides a strong foundation for
follow-up studies investigating the mechanistic basis of the enhanced virulence phenotype, though
mutations that are found to enhance virulence in model systems may not translate to increased virulence
during human infections.
Several alternative approaches can also be used to study pathogenicity. Two types of comparative
sequence analysis can provide insight into viral genetic traits that may contribute to virulence. First,

comparative sequencing of SARS-CoV and MERS-CoV epidemic strains with varying levels of virulence
can lead to the identification of mutations associated with enhanced virulence. A strength of this approach
relative to serial passaging is that comparative sequence analysis uncovers genetic variation that is
specially associated with enhanced virulence in humans.1207,1208 However, this approach is limited to
CoVs that have already produced epidemics in humans, i.e. SARS-CoV and MERS-CoV. The success of
this approach depends on the availability of a wide breadth of surveillance data accompanied by
epidemiological data about the clinical severity and case fatality rates of particular strains or groups or
strains. In addition, the fact that SARS and MERS preferentially cause severe disease in patients who are
elderly, immunocompromised, and/or who suffer from co-morbidities complicates the interpretation of
genetic surveillance data. Because disease pathology can be exacerbated by host factors, such as age, as
well as viral factors, high-quality “metadata” about relevant host factors (e.g. age, immune status, pre-
existing medical conditions, etc.) is needed to control for host factors so that sequences can be
appropriately “binned” into low- and high-virulence categories for comparison.1209,1210,1211 While SARS-
CoV strains from the early, middle, and late phases of the 2002 – 2003 epidemic have been found to
exhibit varying levels of virulence (and have been used for comparative sequence analysis studies),
genetic surveillance data for MERS are limited. Finally, given the large size of the CoV genome and
genetic diversity among wild type CoV sequences, sequence comparisons are practically limited to pre-
determined regions of interest, which precludes identification of novel virulence factors.
A second sequence-based approach involves analyzing the evolution of CoVs over time. Understanding
which regions of the genome mutate and which do not can provide insight into which regions are likely to
be critical for the virus life cycle. Although these regions/factors may not be involved in virulence per se,
this approach may be useful for identifying promising therapeutic targets. However, the utility of this
approach is also limited by the number of available CoV sequences.
Loss-of-function (LoF) studies, which involve knocking out or otherwise hampering the function of a
gene of interest (or its product) and screening for attenuated fitness (in vitro) or virulence (in vivo),
represent another alternative approach for the discovery of viral virulence factors and genetic traits
associated with virulence. Though this approach enables the identification of novel virus proteins that are
necessary for enhanced fitness/virulence, the simple discovery of a novel virulence factor does not
provide information about its potential function. Conversely, a random mutagenesis approach may
provide insight into the mechanistic basis of virulence but is highly inefficient because of the number of
potential targets in the CoV genome. The major drawback of LoF screens is that losing the functionality
of a virus protein, either through gene knockout of mutagenesis, may indirectly attenuate virulence, so
that gaining meaningful information about virulence mechanisms may be difficult using this approach.
One strategy for identifying potentially interesting gene targets for LoF studies is to examine CoV
sequences for the presence of conserved enzymatic motifs. However, a limited number of CoV enzymes
contain recognizable motifs (e.g. the RNA-dependent RNA polymerase), and CoV accessory proteins are
distinctive among CoVs and distinctive in nature.1212 Thus, a LoF approach that relies on targeted
mutagenesis is primarily limited to the investigation of virulence-enhancing mutations in known virulence
1207 Qu XX et al (2005) Identification of two critical amino acid residues of the severe acute respiratory syndrome coronavirus
spike protein for its variation in zoonotic tropism transition via a double substitution strategy. J Biol Chem 280: 29588-
29595
1208 Chinese SMEC (2004) Molecular evolution of the SARS coronavirus during the course of the SARS epidemic in China.
Science 303: 1666-1669
1209 Roberts A et al (2007) A mouse-adapted SARS-coronavirus causes disease and mortality in BALB/c mice. PLoS pathogens
3: e5
1210 Peiris JS et al (2003) Clinical progression and viral load in a community outbreak of coronavirus-associated SARS
pneumonia: a prospective study. Lancet 361: 1767-1772
1211 Assiri A et al (2013) Epidemiological, demographic, and clinical characteristics of 47 cases of Middle East respiratory

factors. For both LoF strategies, a limited number of mutants can be screened for attenuated virulence in
vivo, due to the labor, expense, and ethical considerations associated with the conduct of animal
experiments. Though high-throughput screening for reduced replicative fitness can be conducted using
cell culture systems, as discussed above, replicative fitness does not necessarily correlate with virulence
and represents only one of the phenotypes underlying virulence. Thus, in vitro LoF screening approaches
may lead to false negative and false positive results and can only target a fraction of the virulence factor
space. Finally, it is noted that knocking out the function of an unknown viral protein can lead to a loss or
gain of virulence, depending on the function of the protein. Notably, even genetic manipulations that are
predicted to attenuate virulence based on preliminary in vitro work can lead to enhanced virulence when
tested in an in vivo model system.1213
LoF approaches can also be used to confirm that a particular trait is necessary for enhanced virulence.
However, because virulence is a complex, multi-genic trait, knocking out the function of one gene or
introducing a mutation into one gene may be sufficient to attenuate virulence but provides an incomplete
picture of the role of that particular protein. As above, mutations that are found to enhance virulence in
model systems may not translate to increased virulence during human infections.
section, with respect to the ability of each approach to provide insight into the mechanisms underlying
CoV virulence, are summarized in Table 15.2. Taken together, serial passaging for the selection of CoV
strains with enhanced pathogenicity in animals or fitness in cell culture, a GoF approach, is the most
efficient and effective method for identifying novel genetic traits and/or viral factors that contribute to
virulence in any coronavirus strain. However, results in cell culture or animal model systems may not
translate to human disease. The alternate approaches have several drawbacks. While screening gene
knockout viruses in vitro represents a viable approach for the discovery of novel virulence factors, this
LoF approach is limited to the identification of proteins that influence replicative fitness, only one
component of virulence, and may uncover factors that attenuate virulence for trivial reasons. The main
drawback of both the GoF and LoF approaches is that insights gleaned from model systems may not
translate to human infection. To that end, comparatively analyzing the sequences of SARS/MERS strains
with varied levels of virulence can provide direct insight into genetic traits that are associated with
pathogenicity in humans. However, this approach is limited to the study of SARS and MERS and is
significantly constrained by shortcomings in the quality and availability of existing genetic surveillance
data. In addition, any hypothesis generated through comparative sequence analysis must be
experimentally confirmed. The phenotypic consequences of mutations that are associated with enhanced
virulence can be validated using GoF approaches, which are uniquely capable of demonstrating that
mutations are necessary and sufficient to enhance virulence, or LoF approaches, which can demonstrate
that mutations are necessary for enhanced virulence only. Complex, multi-genic traits such as virulence
are difficult to tease apart using solely LoF approaches because LoF provides limited information about
how proteins cooperate to enhance virulence. However, because the value of the information gleaned
from both LoF and GoF approaches depends on the relevance of artificially manipulated viruses to nature,
using both approaches to confirm the role of a particular mutation or phenotype strengthens any
conclusion.
1213
Eckerle LD et al (2007) High fidelity of murine hepatitis virus replication is decreased in nsp14 exoribonuclease mutants.

Table 15.2. Comparison of GoF approaches and corresponding alt-GoF approaches that benefit scientific knowledge: How do CoVs cause disease?
What are the critical viral virulence factors and viral genetic determinants of virulence?
to enhance fitness must be experimentally
confirmed
• Identify novel genetic traits that are • Simplicity of model system – replicative
sufficient to enhance fitness in cell fitness is one component of virulence and does
GoF #1 [2]: (In vitro approach) Serial passaging of virus in
culture, for any virus not necessarily correlate with virulence in vivo
cells
• Identify novel viral factors that may • Translatability – results from model systems
contribute to virulence, for any virus may not translate to human infections
other CoV strains

• Identify novel genetic traits that are
to enhance virulence must be experimentally
sufficient to enhance virulence, for any
confirmed
GoF #2 [2]: (In vivo approach) Serial passaging of virus in virus
• Translatability – results from model systems
animals • Identify novel viral factors that may
may not translate to human infections
contribute to virulence, for any virus
other CoV strains
• Identify genetic traits that are necessary
and sufficient to enhance
GoF #3 [3]: Targeted genetic modification of virus to
fitness/virulence • Translatability – results from model systems
introduce mutation(s) shown to be associated with enhanced
• Confirm viral factors that contribute to may not translate to human infections
fitness/virulence
virulence • Narrow breadth – results may not generalize to
• Characterize fitness/virulence of mutant in cell culture • Gain insight into mechanisms other CoV strains
or animal model systems
underlying pathogenesis, including
identification of underlying phenotypes

• Utility and success of approach is constrained
by the quality and availability of genetic
surveillance data
• Host factors such as age complicate
interpretation of virulence data
Alt-GoF #1 [1]: (virus free) Comparative sequence analysis • Identify genetic traits that are associated
• Bias – limited to investigation of known
of human epidemic CoV strains with varying levels of with enhanced virulence in humans
genetic regions of interest
virulence • Identify conserved traits, if large
• Reactive – limited to the study of CoVs that
numbers of sequences are analyzed
have already caused human infections (i.e.
SARS and MERS)
and sufficient for adaptation must be
experimentally confirmed
• Predictive – whether regions contribute to

Alt-GoF #2 [2]: (virus free) Comparative analysis of CoV • Identify genetic regions that may be virulence must be experimentally confirmed
sequences over time, to which regions of the genome critical for the virus life cycle
mutate • Utility and success of approach is constrained
by the quality and availability of genetic
surveillance data

• Triviality – losing the function of a virus
protein may indirectly attenuate virulence
• Bias - Targeted mutagenesis strategies
primarily limited to the investigation of known
Alt-GoF #3 [3]: (Loss of function) Forward genetic screen
virulence factors
to identify mutations expected to attenuate replication in
• Inefficient – limited number of mutants can be
vitro or virulence in vivo
• Identify genetic traits or viral factors screened in vivo
• Random mutagenesis of known virulence factors to that are necessary for enhanced • Simplicity of in vitro model system –
generate libraries of mutant viruses, followed by virulence replicative fitness does not necessarily
screening of mutants for attenuated replication/virulence correlate with virulence
• Knock out function of individual genes and screen for • Unpredictable - knocking out the function of a
attenuated replication/virulence protein can lead to a gain or loss of virulence
• Mutations that are necessary for enhanced
virulence may not be sufficient to enhance
virulence in a different genetic context

15.1.4 Benefits of GoF-derived Model Systems
Model systems that can be efficiently infected by CoVs, support robust viral replication, and mimic
human disease pathogenesis are essential for the experimental study of CoV biology and for the
development of MCMs. GoF approaches that expand the host range of CoVs are used for the
development of in vitro and in vivo model systems for SARS, MERS, and animal-origin CoVs (e.g.
SARS/MERS-like bat CoVs, civet CoVs, etc.)
15.1.4.1 GoF benefits to the development of in vitro model systems
Cell culture systems that can be infected and support robust replication of CoVs are essential for the
generation of viral stocks that are used for in vitro and in vivo experiments and for investigating basic
mechanisms of CoV infection using cell biological methods. Both SARS and MERS readily and
persistently infect human cell lines, but many animal CoVs cannot be cultured in vitro, including
SARS/MERS-like bat CoVs and zoonotic SARS strains from civets.1214,1215 Specifically, many animal
CoVs do not naturally infect human cell lines, and some bat CoVs cannot be isolated in bat cell lines
either. Even for those bat CoVs that are capable of naturally infecting bat cells, adaptation to standard
mammalian cell culture systems is desirable because bat cells are more difficult to culture and to
manipulate experimentally (e.g. transfect, etc.) than human cell systems.1216,1217 Therefore, new in vitro
model systems for animal CoVs are needed in order to effectively study the properties of these
SARS/MERS progenitor viruses and to assess their potential to adapt to humans.
15.1.4.1.1 In vitro model systems developed using GoF approaches
Two GoF approaches can be used to adapt animal CoVs for growth in human cells: serial passaging in
cell culture and “Spike swapping.” First, serial passaging in cell culture selects for viruses that are better
able to bind, infect, and replicate within human cells. This approach may not be successful if the initial
capacity of the virus to infect human cells is very low. For example, Becker and colleagues were unable
to recover and passage a consensus bat SARS-like CoV (Bat-SCoV, constructed from four bat SARS-like
CoV sequences) in human cells.1218 The main limitation of this approach is that serial passaging may lead
to the acquisition of mutations that alter the biological behavior of the virus in unexpected ways, which
may limit the relevance of any results to the wild type virus.
A second approach involves “Spike swapping,” targeted genetic modification to replace all or part of an
animal CoV Spike protein with the SARS Spike protein to generate a recombinant chimeric virus (i.e.
animal CoV + SARS Spike). Because the Spike protein is a major determinant of host tropism, this
replacement often enables the chimeric animal-SARS virus to infect and replicate within human
human airway epithelium. Ibid. 82: 2274-2285
1215 Agnihothram S et al (2014) A mouse model for Betacoronavirus subgroup 2c using a bat coronavirus strain HKU5 variant.
mBio 5: e00047-00014
1216 Huynh J et al (2012) Evidence supporting a zoonotic origin of human coronavirus strain NL63. Journal of virology 86:
12816-12825
1218 Becker MM et al (2008) Synthetic recombinant bat SARS-like coronavirus is infectious in cultured cells and in mice.

cells.1219,1220 Of note, chimeric viruses can also serve as a starting point for the development of a mouse-
adapted strain, discussed in more detail in the subsequent section.1221,1222,1223 One benefit of this approach
is that, because the SARS Spike protein mediates entry into a variety of cell types, chimeric viruses can
likely be used to infect both immortalized and primary cell lines (such as human airway epithelial cells,
the site of primary infection of coronaviruses and as such a more relevant model system for the study of
CoV infection than immortalized cell lines).1224 The main drawbacks of this approach are that the
behavior of the chimeric virus may not reflect that of the wild type virus and that chimeric viruses cannot
be used to study the function of the animal Spike protein.
15.1.4.1.2 Alternative in vitro model systems that do not involve the use of GoF approaches
Several alternative model systems, which do not involve GoF approaches, may permit the study of animal
CoVs in cell culture: use of cell lines derived from the natural host (e.g. bat), use of cell lines that are
naturally permissive to infection, and the development of human cell lines that are sensitized to infection
with animal CoVs. First, some bat CoVs are naturally capable of replicating within bat cell lines, such as
a bat SARS-like CoV isolated in 2013 which is thought to be a progenitor strain for SARS.1225 However,
there are several limitations associated with the use of bat cell lines. Bat cell lines are much less
experimentally tractable than human cell lines, as fewer reagents are available and the cells are more
difficult to transfect than human cells.1226,1227 Also, some bat CoVs do not infect existing immortalized bat
cell lines (only a few are available), which restricts their utility as a model system for emerging
CoVs.1228,1229
A second alternative involves the use of naturally permissive cell lines. For example, the SARS-like bat
CoV strain described above was found to naturally replicate in Vero cells (derived from African green
monkeys), human alveolar basal epithelial cells and pig kidney cells.1230 Interestingly, this strain
replicated to higher titers in Vero cells than in bat kidney cells, demonstrating that cells derived from a
natural host species do not necessarily represent a superior model system than cells derived from a non-
natural host species. However, many bat CoVs, such as the MERS-like virus HKU5, cannot be cultured in
standard in vitro systems, limiting the utility of this approach.1231 In addition, CoVs that are found to
1219 For example, swapping the Spike ectodomain from SARS into the backbone of Bat-SCoV permitted replication of the
chimeric virus in a variety of human cell types.
1221 In this case, the Spike protein from the mouse-adapted SARS strain (which contains one amino acid substitution relative to
the WT SARS protein) is used to generate the chimeric virus.
mBio 5: e00047-00014
1224 Dijkman R et al (2013) Isolation and characterization of current human coronavirus strains in primary human epithelial cell
cultures reveal differences in target cell tropism. Journal of virology 87: 6081-6090
535-538
Coronavirus. Journal of virology 89: 9119-9123
1227 Huynh J et al (2012) Evidence supporting a zoonotic origin of human coronavirus strain NL63. Ibid. 86: 12816-12825
1228 ibid.
mBio 5: e00047-00014
535-538
mBio 5: e00047-00014

naturally infect and replicate within bat cells or human cells may yield progeny virus incapable of further
infection.1232
A final alternative involves sensitizing host cells to infection through ectopic expression of the receptor
protein from the natural host species (or another permissive host species). For example, Ge and
colleagues demonstrated that the bat SARS-like CoV described above is capable of infecting and
replicating within HeLa cells expressing the ACE2 receptor from civets or bats, demonstrating the
potential utility of this strategy for development of an in vitro model system for the study of bat CoVs.1233
As this system does not account for host factors governing viral entry and replication other than the host
receptor, whether this strategy will permit replication of a broad range of emerging CoVs is unknown.
Additionally, this strategy cannot be used for primary cell lines, which are not readily transfectable, and
overexpression of the receptor may alter the process of infection, leading to artefactual results. Finally,
within each alternative system, wild type viruses may not replicate to high enough titers for experimental
use without serial passaging to select for higher-yield viruses.
15.1.4.1.3 Summary – benefits of GoF-derived in vitro model systems relative to alternative model
systems
The strengths and limitations of each in vitro model system analyzed in this section are summarized in
Table 15.3. Studying SARS/MERS-like animal CoVs, thought to be precursors for SARS/MERS or to
have similar potential to spill over into human populations, provides important insight into how SARS
and MERS emerged from their animal reservoirs to infect humans. In addition, defining which animal
CoVs have potential to adapt to humans can guide efforts to develop broad-spectrum MCMs for emerging
CoVs. However, most animal CoVs grow poorly, if at all, in standard cell culture systems. GoF
approaches have unique potential to enable the development of in vitro model systems for the study of
any animal CoV in a variety of cell types, including immortalized cell lines and relevant primary cell lines
such as human epithelial airway cells. Alternatives to GoF have significant shortcomings. Only a subset
of animal CoVs identified to date can be cultured in bat, human, or other standard cell lines, limiting the
utility of using naturally permissive cell lines for in vitro studies. While ectopic expression of permissive
receptor proteins in a common cell line has been shown to permit replication of several CoVs, this
strategy is limited to cell lines that can be readily transfected (i.e. not primary cell lines) and
overexpression of the host receptor may alter the biology of infection, limiting the relevance of results
from this system.
535-538

Table 15.3. CoVs: Summary of the Benefits of GoF Approaches that Inform the Development of in vitro model systems
Scientific Knowledge and MCM Development Benefits – Development of in vitro model systems for SARS-CoV and MERS-CoV
Model system Benefits Limitations
GoF: Animal CoV adapted for growth in human
cell lines
• Can be applied to any animal CoV
• Adapted viruses can be used to infect a variety of • The behavior of adapted viruses may not
• Serial passage of animal CoV in human cell cell types, including immortalized and primary reflect that of wild type viruses
lines [1, 4a] cells • Chimeric viruses cannot be used to study the
• Targeted genetic modification to generate • A wide variety of methods and reagents are function of the animal CoV Spike protein
chimeric virus: animal CoV plus SARS Spike available for human cell lines
[5]
• Few bat CoVs infect existing immortalized
bat cell lines (few cell lines are available)
Alt-GoF #1: Use of cell lines derived from the o Progeny may be incapable of further
natural host (e.g. bat) • Enables the use of wild type bat CoVs infecting cells
• Bat cell lines are less experimentally tractable
than human cell lines
• Few bat CoVs can be cultured in standard in
Alt-GoF #2: Use of naturally permissive cell
• Enables the use of wild type animal CoVs vitro systems
lines • Wild type viruses may replicate to higher titers than o Progeny may be incapable of further
in cells derived from natural host infecting cells
• Additional host factors play a role in virus
entry and replication
Alt-GoF #3: Use of human cells that have been o Strategy may not be successful for all
sensitized to infection
• Enables the use of wild type animal CoVs animal CoVs
• A wide variety of methods and reagents are • Overexpression of the receptor may alter
• Ectopic expression of virus receptor from the available for human cell lines
natural host species infection processes
• Limited to the use of host cell lines that can be
readily transfected

15.1.4.2 GoF benefits to the development of animal model systems
Animal models are essential for understanding the pathology of viral disease and for developing vaccines
and therapeutics. Replication models are animals that support viral replication but do not mimic human
disease, while pathogenesis models are those that support viral replication and emulate human
pathologies. If suitable laboratory animals are not naturally susceptible to infection, animal models can be
developed by adapting a wild type virus to the host through passaging or by adapting the host to the virus
by transgenic expression of host viral receptors or other restriction factors. Adapted strains, transgenic
animals, and naturally susceptible species have all been used to study SARS-CoV and MERS-CoV.
Appropriate animal models are critical for the development of new vaccines and therapeutics. To study
vaccine efficacy, the model must show the ability of the vaccine to prevent pathology associated with
infection following a challenge.1234 In addition, under the FDA's Animal Efficacy Rule, vaccines and
therapeutics against rare, emerging, or virulent agents such as SARS-CoV can achieve regulatory
approval provided efficacy is demonstrated in multiple animal models that display clinical illness
representative of human disease.1235,1236 (Whether the Animal Rule applies to the development of MCMs
targeting MERS-CoVs is uncertain, as the number and distribution of MERS cases in the Kingdom of
Saudi Arabia may enable the conduct of clinical trials, which is preferable. This issue will be addressed
on a case-by-case basis if sponsors seek approval of a MERS-CoV vaccine or therapeutic under the
Animal Rule.)1237 In the event that a sponsor seeks approval of a SARS-CoV vaccine or therapeutic under
the Animal Rule, the sponsor must provide scientific justification that the animal used to study
countermeasures exhibits key characteristics of human disease when exposed to the challenge agent and
accurately predicts human responses. In sum, development of a pathogenesis model that adequately
mirrors the route of infection, severity, clinical signs, and levels of mortality and morbidity seen in
humans is critical for advancing countermeasure development and for satisfying the FDA Animal Rule.
15.1.4.2.1 Animal model systems developed using GoF approaches
Virus strains that have been adapted to laboratory animals
Adaptation of a virus to a mammalian host through serial passaging is a commonly used method for
creating pathogenesis models. Because this method results in an additional capability for the virus to
infect and cause disease in a new host species (i.e. altered host range and enhanced pathogenicity in
appropriate animal model systems), this method represents a GoF approach. As neither SARS-CoV nor
MERS-CoV are capable of productively infecting mice to recapitulate human disease pathogenesis,
mouse-adapted strains of SARS-CoV are preferred relative to use of the WT strain, and efforts to
development a mouse-adapted MERS-CoV strain are ongoing.1238 Mouse-adapted strains are important
tools for the study of viral pathogenesis and of host factors involved in responses to infection. Use of
mouse-adapted strains allows researchers to capitalize on the diversity of mouse-specific tools developed
for the study of host immune responses, including many strains of knockout mice and reagents for
manipulating host immune factors (e.g. antibodies for depletion of particular types of host immune cells).
Lessons learned using mouse-adapted strains are likely to be applicable to humans because pathogenesis

1235 Sutton TC, Subbarao K (2015) Development of animal models against emerging coronaviruses: From SARS to MERS
coronavirus. Virology 479-480C: 247-258
1236 FDA. Product Development Under the Animal Rule: Guidance for Industry.
http://www.fda.gov/downloads/drugs/guidancecomplianceregulatoryinformation/guidances/ucm399217.pdf. Last Update
October 2015. Accessed November 23, 2015.
1237 (2015m) Personal communication from FDA representative.

mechanisms, specifically virus entry mechanisms, are similar to those in humans.1239 However, there is
the possibility that unforeseen changes will arise during passaging that may affect pathogenesis and
potentially complicate comparisons between mice and humans.1240 Understanding how adaptation
mechanisms alter the phenotypes under study is critical for the correct interpretation of results.1241 In
addition, adapted strains can be used to test whether candidate MCMs can prevent or reduce the
pathology associated with human disease, which is important for advancing countermeasure development.
The adapted strains of SARS-CoV have been used in vaccine development, representing a significant
advance towards satisfying the FDA Animal Rule.1242
15.1.4.2.2 Alternative animal model systems that do not involve the use of GoF approaches
Transgenic laboratory animals that have been sensitized to infection
Use of transgenic animals expressing the human virus receptor is an alternative to the use of adapted
viruses for hosts that are not permissive or do not recapitulate human disease pathology. Transgenic
approaches have been used to develop mouse models for SARS-CoV and MERS-CoV. Transgenic
models allow the direct study of wild type viruses, thus avoiding the concern that adaptive changes during
passaging alter mechanisms of viral pathogenesis. Transgenic mice are important in countermeasure
development because they can be used to establish that a therapy knocks down virus titers in a system
with human receptors.1243 For MERS-CoV, a transgenic approach can be used as a starting point for the
creation of an adapted strain because mice do not naturally express the appropriate viral entry
receptors.1244 A variety of approaches have been used to create transgenic mouse models for SARS-CoV
and MERS-CoV infection, but each technique results in a slightly different gene expression pattern and
reproduces human disease symptoms to a different degree. As a result, the relevance of results about
pathogenesis mechanisms and MCM efficacy is subject to significant caveats. Notably, to date, no animal
model that includes a genetically modified host has been used to approve an FDA-regulated
countermeasure under the Animal Rule.1245
Naturally Susceptible Species
Another alternative to the use of viruses that have been adapted to laboratory animals is the use of
naturally susceptible hosts. However, laboratory animals that are naturally susceptible to infection with
SARS-CoV and MERS-CoV have been found to support viral replication but remain asymptomatic or
develop symptoms dissimilar to those in humans. SARS-CoV is capable of productively infecting mice,
hamsters, ferrets, and several species of non-human primate, though not all species exhibit clinical signs
or mortality. MERS-CoV, which utilizes a different entry receptor than SARS-CoV, exhibits a greater
degree of host species restriction; replication is limited to some species of non-human primate and no
small mammals are permissive to infection. Thus, for both SARS-CoV and MERS-CoV, naturally
susceptible hosts function as replication models, not pathogenesis models.1246
1239 Ibid.
1240 Frieman, M., et al. (2012). "Molecular determinants of severe acute respiratory syndrome coronavirus pathogenesis and
virulence in young and aged mouse models of human disease." J Virol 86(2): 884-897.
1242 Sutton TC, Subbarao K (2015) Development of animal models against emerging coronaviruses: From SARS to MERS
coronavirus. Virology 479-480C: 247-258.
1244 Ibid.

Replication models are used in vaccine and anti-viral drug development to demonstrate diminished
replication, an important proof of concept for medical countermeasures.1247 Additionally, identifying a
natural replication model is often the first step in creating a pathogenesis model. However, animals that
do not recapitulate human disease pathogenesis have limited utility for investigating how viruses interact
with host systems to cause disease, and asymptomatic replication models do not provide insights into
pathogenesis or disease progression. Replication models also have limited utility for advanced MCM
development. Replication models may provide easy metrics to demonstrate vaccine or drug efficacy (i.e.
reduction in viral replication), but their lack of relevant symptomology could lead to the development and
release of subpar or dangerous countermeasures.1248 Specifically, therapeutics may cause unintended side
effects or deleterious interactions with the host immune system, which are unpredictable and may not be
observed in asymptomatic animal models.1249 This concern is supported by the example of a SARS-CoV
vaccine candidate, which was efficacious in non-human primate replication models but produced severe
adverse side effects when tested in mouse pathogenesis models. After vaccinated mice were challenged
with live SARS-CoV virus, the mice displayed an immunopathologic Th2-type response, which is
predictive of a harmful response to the vaccine in humans.1250 As a result, this vaccine candidate did not
undergo clinical trials.
Alternative coronaviruses that are naturally pathogenic to laboratory animals - Mouse Hepatitis Virus
The coronavirus mouse hepatitis virus (MHV) has been used as a model to generate basic knowledge
about coronavirus biology but cannot serve as a substitute for MERS-CoV or SARS-CoV for
pathogenesis studies or MCM development studies. Adult mouse infections of MHV are usually
asymptomatic. While infant mice exhibit pathology during infection, the symptoms and disease course do
not mimic those of MERS-CoV or SARS-CoV. MHV has been useful for the study of mechanisms
universal to coronaviruses, which has led to the discovery of generalizable information about coronavirus
polymerases, proteases, and other nonstructural proteins.1251 However, coronaviruses do not share the core
machinery often targeted by antivirals or vaccines, and studies have shown that inhibitors that
successfully target one coronavirus do not work for the other.1252,1253 Thus, the efficacy of all
countermeasures tested in the context of MHV infection must be confirmed using SARS-CoV or MERS-
CoV. In addition, due to the unique features of SARS-CoV and MERS-CoV, pathogenesis,
transmissibility, and the effects of SARS-CoV and MERS-CoV in humans cannot be studied using
MHV.1254
Human Autopsy Data
Human autopsy data can be an alternative source of pathogenesis information. SARS associated lung
pathology was described from examination of post-mortem tissue samples; however, pathologic changes
associated with MERS have not been reported due to a lack of autopsy data.1255 Autopsies are not often
performed in Middle Eastern cultures, and data has not yet been shared from the most recent outbreak in
1247 Ibid.
1248 Ibid.
1249 Ibid.
1250 Tseng, C. T., et al. (2012). "Immunization with SARS-CoV coronavirus vaccines leads to pulmonary immunopathology on
challenge with the SARS-CoV virus." PLoS One 7(4): e35421.
1252 Ibid.
1253 Hilgenfeld R (2014) From SARS to MERS: crystallographic studies on coronaviral proteases enable antiviral drug design.
The FEBS journal 281: 4085-4096
1255 Gretebeck LM, Subbarao K (2015) Animal models for SARS and MERS coronaviruses. Current opinion in virology 13:
123-129.

the Republic of Korea.1256 Human autopsy data is inherently correlative and is devoid of time series
information, obscuring the order in which pathogenic effects occurred. Diversity in genetic backgrounds,
life histories, and chronic conditions must all be taken into account and can complicate the identification
of pathology caused by viral infection versus comorbidities. MERS-CoV has increased mortality rates in
the elderly and those with pre-existing health conditions, so information from these individuals may not
fully represent pathology seen in younger, otherwise healthy persons.
15.1.4.2.3 Summary – benefits of GoF-derived in vitro model systems relative to alternative model
systems
Model systems are essential for understanding the pathology of viral disease and for developing vaccines
and therapeutics. The strengths and limitations of each in vivo model system analyzed in this section are
summarized in Table 15.4. Mouse-adapted strains of SARS, which exhibit altered host range and
enhanced virulence in mice relative to the wild type SARS virus, represent the only model system that
recapitulates disease pathogenesis observed during human infections of SARS-CoV. As existing animal
models for MERS-CoV do not replicate human disease pathology, mouse-adapted strains of MERS-CoV
are expected to serve as the sole pathogenesis model for the study of MERS-CoV infection as well. As
such, animal-adapted strains can be used to study many facets of disease pathogenesis, including the
course of disease, the role of viral and host immune factors in disease pathology, and the role tissue
tropism in disease pathology. Alternative model systems have critical drawbacks for the study of disease
pathogenesis. Transgenic animals do not recapitulate the features of human disease because the
engineered animals do not exhibit native expression patterns of viral receptor proteins. As a result, lessons
learned about pathogenesis may not translate to humans, and transgenic animals cannot be used to study
the role of tissue tropism in disease pathology. Most naturally susceptible hosts are asymptomatic or
display dissimilar symptoms to humans and thus cannot be used to study disease pathogenesis. While
human autopsy data are uniquely capable of providing insight into human disease pathology, limited
autopsy data are available, and the static nature of the data and the presence of co-morbidities in many
SARS/MERS patients complicate interpretation of the data.
The use of animal-adapted strains of CoVs is critical for advanced MCM development as well and
provides significant advantages over the use of alternative model systems. Though transgenic animals and
naturally susceptible hosts can be used to demonstrate that MCMs diminish viral replication, an important
proof of concept for early stage MCMs, animal-adapted strains that replicate human disease pathology
provide a much more robust system for demonstrating the safety and efficacy of MCM candidates. In
addition, because adapted strains provoke a response from the host immune system, use of these strains
can reveal MCM side effects or adverse reactions that are not seen in asymptomatic models.

Table 15.4. CoVs: Summary of the Benefits of GoF Approaches that Alter Host Tropism and Enhance Virulence
Scientific Knowledge and MCM Development Benefits – Development of in vivo model systems for SARS-CoV and MERS-CoV
• Animal-adapted strains recapitulate the
GoF [4b]: Animal-adapted SARS-CoV/MERS- pathology of human disease
• Mutations that arise during passaging may alter
CoV o Suitable for the study of disease
pathogenesis in unexpected ways, complicating
• Serial passage of SARS/MERS virus in animals pathogenesis mechanisms
comparisons between mice and humans
(e.g. mice) o Robust system for testing the safety and
efficacy of MCMs
• Naturally susceptible hosts are asymptomatic or
display different symptoms than humans
• Enables the use of wild type virus strains o Cannot be used for the study of pathogenesis
Alt-GoF #1: Use of naturally susceptible
• Can be used to demonstrate that MCMs mechanisms
laboratory animal hosts
diminish viral replication o Weak system for testing the efficacy of MCMs
o Do not display adverse reactions/side effects of
MCMs
Alt-GoF #2: Use of transgenic animals: sensitize
non-permissive host to infection through
• Transgenic animals not mimic human pathogenesis
due to different transgene expression patterns than in
expression of virus entry receptor • Enables the use of wild type virus strains
humans
• Stable expression of human receptor (e.g. • Can be used to demonstrate that MCMs o Cannot be used to investigate tissue tropism, and
knock-in mouse) using universal or host diminish viral replication in a system with pathogenesis mechanisms may not translate to
promoter human virus receptors humans
• Transient expression of human receptor (e.g. o MCM testing results may not translate to humans
adenovirus vector-based transduction)
• Data limited by infrequency of autopsies in Middle
East
Alt-GoF #3: Use of human autopsy data from • Provides direct information about human • Mortalities are not representative of all cases
MERS-CoV cases pathology o Higher incidence of mortality in patients with co-
morbidities

Table 15.4. CoVs: Summary of the Benefits of GoF Approaches that Alter Host Tropism and Enhance Virulence
Scientific Knowledge and MCM Development Benefits – Development of in vivo model systems for SARS-CoV and MERS-CoV
• Does not replicate human disease pathogenesis
Alt-GoF #4: Use of alternative coronavirus: • Can be used to gain insight into basic • Does not share core machinery often targeted by
Mouse Hepatitis Virus (MHV) aspects of coronavirus biology MCMs with SARS or MERS

15.1.5 Benefits of GoF to Public Health/Medicine
GoF approaches have potential to benefit the development of vaccines and therapeutics for coronaviruses
in two ways. First, scientific information gleaned using GoF approaches may inform the development of
new medical countermeasures and supports their licensure. Second, model systems developed using GoF
approaches can be used to demonstrate the safety and efficacy of candidate vaccines and therapeutics.
This section evaluates the benefits of both types of GoF approaches for the development of vaccines and
therapeutics, relative to alternative experimental approaches as well as alternative scientific and technical
innovations that have the potential to similarly benefit MCM development.
For both vaccines and therapeutics, several different types of GoF research (i.e. different GoF
phenotypes) can inform different stages of the MCM development and licensure process. To promote an
understanding of the criticality of GoF approaches for the creation of new vaccines and therapeutics, it is
necessary to first evaluate all GoF approaches that contribute to the vaccine development process (which
includes multiple GoF phenotypes), and then evaluate all GoF approaches that contribute to the
development of new therapeutics (which includes multiple GoF phenotypes). Within the vaccine and
therapeutic sub-sections, the process of developing a vaccine or therapeutic, from development to
licensure, is outlined and the role of GoF versus alternative approaches at each stage of the process is
evaluated. (Note that this structure is slightly different from other sub-sections of this chapter, in which all
GoF approaches and all alternative approaches in turn were discussed.) The sub-section concludes with an
assessment of the contribution of GoF approaches to the development of broad-spectrum vaccines and
therapeutics.
15.1.5.1 Development of new coronavirus vaccines
Currently, there are no FDA-approved vaccines for CoVs, which represents a critical gap in our public
health preparedness for CoV outbreaks.
15.1.5.1.1 Developing new vaccine platforms
Live attenuated vaccines developed using GoF approaches
GoF approaches have the potential to benefit two aspects of the development of live attenuated vaccine
(LAV) platforms, which is a type of vaccine that is being actively researched for its potential as a CoV
vaccine platform. First, GoF approaches can inform the development of candidate LAV strains, which
exhibit attenuated virulence relative to parental strains. Specifically, one strategy for generating LAV
strains is through serial passaging in a non-human host (either an animal or cells derived from an animal),
as adapting a virus to a new host typically attenuates the virus in humans (i.e. alters rather than enhances
host tropism). Because this approach alters host tropism, it is considered to be a GoF approach under the
NSABB Framework. Although serial passaging has been used historically for developing polio, smallpox
and other viral vaccines, the approach has not been utilized for the purpose of developing CoV vaccine
strains.1257
Another strategy for developing attenuated vaccine strains is through targeted mutagenesis to attenuate or
knock out the function of known virulence factors. As discussed above, GoF studies seeking to develop
strains with enhanced virulence represent the most efficient and effective strategy for identifying CoV
virulence factors, though LoF approaches may also be used. For the purpose of developing LAV strains,
one benefit of LoF approaches is that the experiment may directly generate an attenuated strain. In
1257
Ulmer JB et al (2006) Vaccine manufacturing: challenges and solutions. Nature biotechnology 24: 1377-1383

contrast, GoF approaches that lead to the identification of virulence factors require follow-up studies to
determine how to attenuate that factor or to render it non-functional. Nonetheless, given that few
virulence factors have been identified in SARS/MERS, GoF methods currently represent the most
efficient and viable approaches to inform the development of LAV strain candidates through targeted
genetic modification.
Once a candidate LAV strain has been generated, the strain is typically serially passaged in vitro or in
vivo to determine whether the virus recovers fitness/virulence, which represents a GoF approach by
enhancing its fitness in culture or virulence in vivo. Because a tendency to revert or acquire
compensatory mutations that enhance fitness/virulence could seriously compromise the safety of a live
attenuated vaccine, demonstrating the genetic stability of a candidate LAV is a critical aspect of its
development. The rationale behind this concern is evidenced by the example of a candidate LAV strain
for SARS, which was attenuated through targeted mutagenesis to disrupt the ion channel activity of the
SARS E protein. Upon passaging in cell culture and in mice, the mutant virus acquired compensatory
mutations that restored both ion channel activity and virulence, highlighting the risks associated with live
attenuated vaccines.1258 There are no alternative approaches that can provide this information.
Live attenuated vaccines are an appealing type of vaccine for CoVs for several reasons, including the fact
that they mimic the natural infection cycle better than other types of vaccines, which may induce stronger
and more protective immune responses, and that they can be administered in the same way that natural
infections are acquired to trigger mucosal immunity, which is difficult to generate but is an important
objective for achieving long-term protection against mucosal pathogens such as CoVs.1259, 1260 Two
different LAV candidates for SARS have been shown to completely protect against lethal virus challenge
in mice, demonstrating the promise of this type of vaccine for CoVs.1261,1262 The main concern associated
with LAVs is their potential to regain virulence in people, especially elderly and immunocompromised
people, who are important target groups for CoV vaccines due to their increased susceptibility to severe
infection.1263
Alternative types of vaccines that do involve GoF for their initial development
Several other types of CoV vaccines are in development, which do not rely on GoF approaches for their
initial development, including inactivated whole virus vaccines, recombinant vaccines, DNA vaccines,
viral vector-based vaccines, and virus-like particles (VLPs).1264 Many of these vaccine types have shown
promise, and each has strengths and limitations relative to the use of live attenuated vaccines. For
example, DNA vaccines, which consist of plasmid DNA that encodes CoV proteins, are safe (because
they do not contain infectious material) and are easy to design, stable, and inexpensive. However, DNA
vaccines generally induce less protective immune responses than inactivated or live attenuated vaccines.
Viral vector-based vaccines, which consist of a different virus (such as adenovirus) expressing a CoV
1258 Nieto-Torres JL et al (2014) Severe acute respiratory syndrome coronavirus envelope protein ion channel activity promotes
virus fitness and pathogenesis. PLoS pathogens 10: e1004077
1264 Ibid.

protein(s), elicit stronger immune responses than DNA vaccines but may cause harmful immune
responses and inflammation.1265,1266,1267
The abilities and limitations of GoF and alt-GoF approaches to support the development of new CoV
vaccines are summarized in Table 15.5. Taken together, both GoF and alt-GoF approaches contribute to
the development of LAVs, and both LAVs and alternative vaccine platforms have shown promise. The
type or types of vaccines that will ultimately prove to be most effective for SARS, MERS, and
SARS/MERS-like coronaviruses is not yet clear based on vaccinology research conducted to date.1268
Given the need for CoV vaccines, pursuing all promising strategies for vaccine development in tandem,
including LAVs, will ensure that an effective vaccine is achieved in the shortest possible period of time.
1265 Weingartl H et al (2004) Immunization with modified vaccinia virus Ankara-based recombinant vaccine against severe
acute respiratory syndrome is associated with enhanced hepatitis in ferrets. Journal of virology 78: 12672-12676
1266 Deming D et al (2006) Vaccine efficacy in senescent mice challenged with recombinant SARS-CoV bearing epidemic and
zoonotic spike variants. PLoS medicine 3: e525
1267 Enjuanes L et al (2008) Vaccines to prevent severe acute respiratory syndrome coronavirus-induced disease. Virus research
133: 45-62

Table 15.5. CoVs: Summary of the Benefits of GoF Approaches that Enhance Virulence in Mammals
Benefits to vaccine development: Develop new candidate vaccines
Approach Benefits Limitations
Support development of LAVs, which have several advantages as a • Cannot demonstrate that mutation
GoF Experimental Approaches: CoV vaccine platform or deletion of a given virulence
GoF Experimental Approaches: • Most efficient and effective strategies for discovering novel viral factor is sufficient to attenuate
• Serial passaging of viruses in cells or virulence traits that are conserved across multiple virus strains, viral replication and/or virulence
animals [2] which may be good targets for attenuation for the development • Concern that LAVs could recover
of candidate LAVs virulence in people necessitates
• Genetic modification to introduce genetic
traits expected to enhance virulence [3] • Determine whether LAV candidates recover virulence upon stringent safety testing
passage in cells/animals, an important aspect of safety testing
Alt-GoF approach #1:
Alternative Experimental Approaches:
• Can be used to demonstrate that mutating or deleting a viral • Limited utility for the discovery of
• Genetic modification to introduce traits novel viral factors that contribute
expected to attenuate virulence (loss of virulence factor is sufficient to attenuate virus replication and/or
to virulence, relative to GoF
function) virulence
approaches
• Comparative sequence analysis of wild
type strains with varied levels of virulence
Alternative vaccine platforms that do not rely
• Many alternative vaccine platforms have shown promise for • Each alternative vaccine platform
on GoF has a unique set of weaknesses
CoV vaccines
• Recombinant vaccines, DNA vaccines , relative to LAVs
and several others

15.1.5.1.2 Evaluating the safety and efficacy of new vaccine candidates
Ultimately, safety and efficacy testing of any vaccine must be conducted in an animal model that
replicates human disease pathogenesis. As discussed above, the use of a pathogenesis model is critical for
safety testing because pathogenesis models can reveal adverse side effects that replication models do not.
Currently, the mouse-adapted SARS virus represents the best pathogenesis model for SARS-CoV
infection. None of the current animal models for MERS replicate human disease pathology, and CoV
researchers believe that adapting the virus for growth in mice, a GoF approach, is the most promising
strategy for developing a pathogenesis model for MERS-CoV infection. Therefore, the development of
animal-adapted viruses using GoF approaches is critical for the development of new CoV vaccines.
15.1.5.1.3 Summary – benefits of GoF to CoV vaccine development, relative to alternative approaches
Taken together, GoF approaches uniquely benefit several aspects of CoV vaccine development. First,
GoF approaches involving the creation of strains with enhanced virulence represent the most efficient and
effective strategy for identifying novel virulence factors to inform the development of candidate live
attenuated vaccine strains, although LoF approaches can also be used and are critical for confirming that
blocking the function of a virulence factor is sufficient to attenuate virulence. Second, GoF approaches
(selecting for enhanced fitness/virulence) are uniquely capable of demonstrating whether LAV strains
recover virulence upon growth in vivo, an important aspect of LAV safety. Finally, animal models
developed using GoF approaches selecting for altered host range and enhanced virulence are critical for
testing the safety and efficacy of any type of vaccine.
15.1.5.2 Development of new coronavirus therapeutics
Currently, there are no FDA-approved therapeutics for CoVs, which represents a critical gap in public
health preparedness for CoV outbreaks.
investigational product, and the genotypic and phenotypic characterization of resistant viruses.1269
Mechanism of action studies should demonstrate the investigational product’s ability to specifically
inhibit viral replication or virus-specific function and should establish the site of the product’s action.
GoF approaches have the potential to directly benefit several aspects of therapeutic development: (1) the
identification of new therapeutic targets, (2) the determination of a drug’s mechanism of action and the in
vitro selection of resistant viruses, to support an IND application, and (3) the determination of dosing
and/or combination therapies that are least likely to lead to emergence of resistance.
15.1.5.2.1 Identifying new therapeutic targets
CoV researchers cited the lack of knowledge of good viral targets for therapeutics as a critical limitation
for the development of CoV therapeutics.1270 As viral virulence factors are potentially good therapeutic

targets, GoF approaches that enhance virulence in cell culture or animal models have the potential to
benefit the development of therapeutics by enabling the identification of new virulence factors. As
discussed above, although alt-GoF approaches such as comparative analysis of the sequences of SARS
epidemic strains or LoF approaches may also lead to the identification of viral proteins that contribute to
virulence, GoF approaches currently represent the most efficient and effective way to identify novel
virulence factors and gain insight into their mechanism of activity, a foundation for the development of
antivirals. Ideally, researchers will identify conserved virulence factors that can be targeted by broad-
spectrum therapeutics or using therapeutic platforms that can be readily adapted for emerging CoVs.
Whether such virulence factors exist is not yet known, and additional research to identify and characterize
the virulence factors of SARS, MERS, and SARS/MERS-like progenitor CoVs is needed to determine the
feasibility of this approach. Notably, LoF approaches are needed to determine whether inhibiting or
attenuating the function of a virulence factor is sufficient to reduce viral replication and/or infection-
associated pathology during infection.
An alternative approach to the targeted development of therapeutics involves high-throughput screening

of compounds for their ability to reduce viral replication in vitro.1271,1272,1273,1274,1275 This is also an active
area of therapeutic research in the CoV field and has generated several promising candidates. One
drawback of this approach is that it is limited to the identification of compounds that reduce viral
replication, which is only one aspect of virulence. Targeting other aspects of virulence, such as viral
interactions with the host immune system, may prove to be a more effective therapeutic strategy. A
related alternative approach involves high-throughput screening of panels of monoclonal antibodies
(mAbs) to identify mAbs that bind to CoV Spike proteins, as mAbs targeting the Spike protein have been
shown to effectively prevent viruses from infecting cells and could prime the immune system to clear the
infection.1276 One potential drawback of this therapeutic strategy is that CoVs can readily acquire
mutations in their Spike protein that enable escape from mAb neutralization; however, researchers are
actively pursuing the development of “cocktails” of mAbs that are more robust to the generation of escape
mutants.1277,1278 Additional drawbacks are that antibody-based therapeutics, which are uncommon for
infectious diseases, may only slow infections and must be injected because antibodies are not small
molecules.
The strengths and weaknesses of GoF and alt-GoF approaches for informing the development of new
CoV therapeutics are summarized in Table 15.6. Taken together, both GoF and alt-GoF approaches
represent promising strategies for the development of candidate therapeutics. The types of therapeutics
that will ultimately prove to be most effective for SARS, MERS, and SARS/MERS-like coronaviruses is
not yet clear based on therapeutic research conducted to date. 1279 Given the need for CoV therapeutics,
1271 de Wilde AH et al (2014) Screening of an FDA-approved compound library identifies four small-molecule inhibitors of
Middle East respiratory syndrome coronavirus replication in cell culture. Antimicrobial agents and chemotherapy 58: 4875-
4884
1272 Dyall J et al ibid.Repurposing of clinically developed drugs for treatment of Middle East respiratory syndrome coronavirus
infection. 4885-4893
1274 Wu CY et al (2004) Small molecules targeting severe acute respiratory syndrome human coronavirus. Ibid. 101: 10012-
10017
1275 Severson WE et al (2007) Development and validation of a high-throughput screen for inhibitors of SARS CoV and its
application in screening of a 100,000-compound library. Journal of biomolecular screening 12: 33-40
1276 Sui J et al (2008) Broadening of neutralization activity to directly block a dominant antibody-driven SARS-coronavirus
evolution pathway. PLoS pathogens 4: e1000197
1278 Sui J et al (2014) Effects of human anti-spike protein receptor binding domain antibodies on severe acute respiratory
syndrome coronavirus neutralization escape and fitness. Journal of virology 88: 13769-13780

pursuing all promising strategies for therapeutic development in tandem will ensure that an effective
vaccine is achieved in the shortest possible period of time.

Table 15.6. CoVs: Summary of the Benefits of GoF Approaches that Enhance Virulence in Mammals
Benefits to therapeutic development: Develop new candidate therapeutics
GoF Approach #1:
GoF Experimental Approaches:
• Most efficient and effective strategies for discovering • Cannot demonstrate that inhibition of a given
• Serial passaging of viruses in cells or novel viral virulence traits that are conserved across
animals [2] virulence factor is sufficient to attenuate
multiple virus strains, which may be good targets for new
disease pathogenesis
• Genetic modification to introduce therapeutics
genetic traits expected to enhance
virulence [3]
Alternative Experimental Approaches:
• Genetic modification to introduce • Can be used to demonstrate that blocking or attenuating • Limited utility for the discovery of novel viral
traits expected to attenuate virulence the function of a viral virulence trait is sufficient to factors that contribute to virulence, relative to
(loss of function) attenuate disease pathogenesis GoF approaches
• Comparative sequence analysis of wild
type strains with varied levels of
virulence
• Limited to the discovery of compounds that
High-throughput screening of small • Approach has generated several promising therapeutic inhibit viral replication, which is only one
molecule compounds to identify those candidates aspect of pathogenesis
that inhibit viral replication in vitro
• CoVs can readily acquire mutations that
Alt-GoF approach #3: confer resistance to neutralization by a given
Identify neutralizing monoclonal • Approach has generated several promising therapeutic mAb
antibodies (mAbs) targeting the CoV candidates • mAb-based therapeutics have several
Spike protein drawbacks, including high production costs
and the need for injection-based delivery

15.1.5.2.2 Determining the mechanism of antiviral activity of a therapeutic
the licensing process.1280 As discussed above, the CoV field is currently pursuing three strategies for drug
development: (1) the deliberate targeting of known virulence factors or virulence pathways, (2) high-
throughput screening of panels of mAbs (either derived from convalescent patient sera or from libraries of
de novo generated mAbs) to identify mAbs that bind to CoV Spike proteins, and (3) high-throughput
screening of FDA-approved drugs to identify therapeutics that inhibit viral replication in vitro. In the first
two cases, the viral target of the therapeutic may be known, whereas in the last case, the target of the
therapeutic is unknown, including whether the therapeutic targets the virus or the host. GoF approaches
can be used to gain insight in the mechanism of activity of a therapeutic, thus benefitting the development
of new drugs. Here the benefit of GoF approaches, relative to alternative experimental approaches, for the
determination of antiviral mechanisms in both of these scenarios is evaluated.
GoF approaches – benefits and limitations
by the therapeutic, or the phenotypic consequences of mutations may be unclear.1281,1282,1283 Additionally,
if the drug targets a host protein, this approach provides indirect information about its mechanism of
activity, which must be inferred based on prior knowledge of virus-host interactions.
Alternative approaches – benefits and limitations
642-646

information about potential viral targets of therapeutics. Viral targets must be inferred based on prior
knowledge of virus-host interactions, which is likely to be challenging given that current knowledge
about CoV-host interactions is limited. Furthermore, because this kind of indirect information does not
provide insight into antiviral mechanisms, this host-focused approach is of limited value for the study of
therapeutics with known viral targets.
a process in the viral life cycle may be difficult from such a static snapshot. In addition, this approach is
less suitable for investigating therapeutics that target a protein-protein or protein-nucleic acid complex
(either a virus-host complex or a virus-virus complex), either to inhibit the function or block the
formation of the complex. The relevant interaction partner may be unknown, or recombinantly producing
and crystallizing the protein complex may be difficult. Critically, because of the high level of effort
required for X-ray crystallography, it is not a feasible approach for simply screening the potential viral
targets of an unknown antiviral.
of the location of the drug binding site.1287 However, inferring the mechanism of antiviral activity based
on knowledge about the drug-virus protein interaction may be difficult, and the approach is less suitable
for studying therapeutics that target a protein-protein or protein-nucleic acid complex (either a virus-host
complex or a virus-virus complex).
Summary – Benefits of GoF approaches relative to alternative approaches
The strengths and limitations of GoF and alt-GoF approaches that can provide insight into the mechanism
of action of a new therapeutic are summarized in table 15.7. Taken together, serial passaging of a virus in
the presence of therapeutic to discover mutations that confer resistance, a GoF approach, is uniquely
capable of identifying the viral target of a novel therapeutic with an unknown mechanism of action. For
therapeutics with known viral targets, this information about resistance mutations can provide
foundational information to guide follow-up structural, cell biological, and biochemical studies
investigating the mechanism of action of the therapeutic. Although crystallography and photoaffinity

cross-linking can also provide insight into the antiviral mechanisms of therapeutics that directly bind to
and inhibit virus proteins, inferring mechanistic information based on static information about the virus-
antiviral complex may be difficult. Finally, the identification of host factors that are required for antiviral
activity is a critical aspect of examining therapeutics with unknown targets. Though solely using host-
focused approaches to elucidate the antiviral mechanism of a therapeutic that targets the virus would be
difficult, this information complements GoF approaches to strengthen the evidence base for the drug’s
mechanism of action.

Table 15.7. Summary of the Benefits of GoF Approaches that Lead to Evasion of Therapeutics
Benefits to therapeutic development: Identify the mechanism of action of a candidate therapeutic
• Identify the viral protein target of a candidate • Elucidating the mechanism of action of a therapeutic based
on indirect information about resistance mutations may be
therapeutic with an unknown target
GoF Approach #1: difficult
Serial passaging of viruses in the
• Provide insight into the mechanism of action o Resistance mutations may arise in non-target proteins,
of the therapeutic through the identification of confounding interpretation of results
presence of therapeutic [8]
mutations that confer resistance
• Not suitable for identifying the targets of therapeutics that
target host proteins
Alt-GoF Approach #1:
RNAi screen targeting host proteins to • Identify the host protein target of a candidate
therapeutic with an unknown target • Provides indirect information about the viral protein targets
identify host proteins that are critical
of a therapeutic
for the antiviral activity of a
therapeutic
• Limited to the study of therapeutics with known targets

Alt-GoF Approach #2: • Provides direct information about the
interaction between a therapeutic and its viral • Inferring mechanism of activity based on static information
Analyze the crystal structure of a protein target about the therapeutic-viral protein interaction may be
therapeutic in complex with its viral difficult
protein target o May provide insight into the mechanism of
antiviral activity • Approach may not be suitable for the study of therapeutics
that target protein-protein protein-nucleic acid complexes

• Provides direct information about the binding
Alt-GoF Approach #3: site of a therapeutic on its viral protein target • Inferring mechanism of activity based on static information
about the therapeutic binding site may be difficult
Photo-affinity crosslinking o May provide insight into the mechanism of

15.1.5.2.3 Determining the genetic threshold for resistance development
resistant viruses and phenotypic characterization of resistant viruses.1288 Selection for resistance studies
should be repeated multiple times to determine if the same or different patterns of resistance mutations
develop, as well as to determine how the concentration of the therapeutic impacts how readily resistance
develops. These studies constitute GoF approaches. The FDA guidance does not suggest any alternative
approaches that could provide similar information. In fact, prior to deployment of the therapeutic and the
emergence of resistant viruses in nature, no alternative approaches can provide this information. Thus,
essential for the licensing of new therapeutics.
15.1.5.2.4 Determining the therapeutic dosage and/or combination therapies that are least likely to lead
to the emergence of resistance
whether and how readily antiviral resistance arises. In the context of candidate CoV therapeutics,
combination therapies are relevant for the development of mAb-based therapeutics. Although mutations
that prevent mAb binding may readily arise in the presence of a single mAb, acquiring mutations that
confer resistance to multiple mAbs that target different sites on a virus protein may be difficult without
compromising viability.1289
administration of multiple therapeutics in combination. First, serial passaging of virus in animals dosed
with varying amounts of the therapeutic provides insight into the dose-dependence of the emergence of
resistant viruses. Because host-dependent factors, such as the rate of metabolism or clearance of the
therapeutic, influence the concentration of therapeutic the virus experiences, conducting passaging studies
in animals provides more relevant information than in vitro passaging studies. Second, serial passaging of
virus in cells or in animals in the presence of multiple mAbs (or other types of therapeutics) can be used
to determine how readily resistance arises in response to combination versus single therapies. Although in
vitro selection studies are useful for screening different combinations of therapeutics, because of the role
of bioavailability and other host-dependent factors on antiviral efficacy, all promising combination
therapies should be validated through in vivo passaging experiments. No alternative approaches are
capable of providing similar information about the dose-dependence of resistance or whether combination
therapies lead to resistance less readily than individual therapies.
Taken together, GoF approaches that lead to the generation of viruses that are resistant to therapeutics in
development are uniquely capable of determining the therapeutic dose that is least likely to lead to the
acquisition of antiviral resistance as well as determining whether combination therapies better prevent the

emergence of resistant viruses than individual therapies. Both types of information benefit the
development of therapeutic strategies that will be effective for a longer period of time in the field.
15.1.5.2.5 Efficacy testing for regulatory approval
Currently, several animal models are available for the testing of SARS-CoV therapeutics: mouse-adapted
strains (GoF), transgenic mice that have been sensitized to SARS infection through expression of the
human ACE2 receptor (alt-GoF), and naturally susceptible species such as mice and ferrets. The mouse-
adapted strains represent the only animal model system for SARS that replicates human disease pathology
and thus provides a much more robust system for demonstrating the safety and efficacy of therapeutic
candidates than other model systems. Additionally, mouse-adapted SARS strain may facilitate the
licensing of therapeutics under the FDA’s Animal Efficacy rule, which states that therapeutics against
rare, emerging, or virulent agents such as SARS-CoV can achieve regulatory approval provided efficacy
is demonstrated in multiple animal models that display clinical illness representative of human disease.1290
Two types of animal models are available for MERS: naturally susceptible hosts, such as rabbits, and
transgenic animals that have been sensitized to MERS infection through expression of the human DPP4
receptor. None of the model systems that have been developed in either category replicate human disease
pathology. Although these systems can be used to demonstrate that MCMs diminish viral replication, the
relevance of results to human disease is uncertain, and these models cannot establish whether a
therapeutic candidate is likely to reduce disease-associated pathology in humans. For that reason,
researchers are actively pursuing the development of a mouse-adapted MERS strain through serial
passaging approaches (GoF), which is thought to be the most promising strategy for developing a
pathogenesis model for MERS-CoV infection that is suitable for advanced MCM testing.
Taken together, GoF approaches, namely serial passaging to develop animal-adapted strains that
recapitulate human disease pathology during infection, are are critical for testing the safety and efficacy
of therapeutic candidates, thereby advancing therapeutic development.
15.1.5.3 Development of broad-spectrum vaccines and therapeutics
Although SARS-CoV is no longer circulating in nature, surveillance efforts over the past decade have
revealed that SARS-CoV and MERS-CoV emerged from a reservoir of thousands of bat CoVs, many of
which are genetically similar to SARS-CoV and MERS-CoV.1291,1292 One SARS-like bat CoV was
recently shown to be naturally capable of infecting human cells, suggesting that SARS/MERS-like bat
CoVs have the potential to spill over into human populations.1293 CoV researchers hypothesize that
additional animal CoVs will emerge to cause epidemics because changing population patterns
increasingly support the ability of CoVs to cause disease and spread in human populations. Namely, both
crowding, which facilitates large respiratory droplet transmission of CoVs, and elderly populations, who
are more susceptible to severe infection and death than younger age groups, are increasing worldwide.1294
For that reason, CoV researchers are strongly interested in developing broad-spectrum vaccines and
therapeutics that will be capable of targeting the next emerging CoV.
1290 Food and Drug Administration. Guidance for Industry: Product Development Under the Animal Rule.
http://www.fda.gov/downloads/drugs/guidancecomplianceregulatoryinformation/guidances/ucm399217.pdf. Last Update
May 2014. Accessed 14 October 2015.
1291 Vijaykrishna D et al (2007) Evolutionary insights into the ecology of coronaviruses. Journal of virology 81: 4012-4020
1292 Graham RL et al (2013) A decade after SARS: strategies for controlling emerging coronaviruses. Nature reviews
Microbiology 11: 836-848
535-538

15.1.5.3.1 GoF approaches – benefits and limitations
The generation of chimeric bat-SARS viruses through recombinant methods (“Spike swapping”),
considered a GoF approach because the host tropism of the chimeric virus may be altered relative to
that of the parental viruses, has the potential to benefit the development of broad-spectrum MCMs.
Specifically, chimeric viruses are used as challenge viruses to explore the broad-spectrum potential of
candidate vaccines and therapeutics, in order to test whether MCMs designed to target SARS/MERS
proteins are also capable of targeting cognate proteins in bat CoVs as well as whether MCMs can target
SARS/MERS proteins in a different virus context (representative of the next emerging CoV capable of
infecting humans). These experiments can provide insight into whether MCMs targeting any CoV protein
or process are capable of conferring broad-spectrum protection against bat CoVs with zoonotic potential,
in addition to SARS and MERS. The major drawback of this approach is that results using artificial
chimeric viruses may not reflect the capacity of MCMs to target the wild type viruses.
15.1.5.3.2 Alt-GoF approaches – benefits and limitations
Several alternative approaches can be used to evaluate the broad-spectrum potential of candidate vaccines
and therapeutics. One approach involves the use of wild type bat CoVs as challenge viruses, in lieu of
chimeric bat-SARS viruses. However, the fact that few bat CoVs can be grown in culture or in animals
without the use of GoF approaches (serial passaging or the generation of chimeric viruses) diminishes the
utility of this approach.
For evaluating vaccines or monoclonal antibody therapies that target the Spike protein, the use of
pseudotyped viruses represents another alternative approach. Because Spike proteins are presented
differently in the context of pseudotyped viruses versus CoVs, especially quaternary epitopes that are
critical for the specificity of Spike-antibody interactions, results using pseudotyped viruses may not be
recapitulated in the context of the wild type virus.1295 For example, researchers reported that certain mAbs
that do not neutralize the wild type SARS virus are capable of neutralizing viruses that are pseudotyped
with SARS Spike proteins.1296 Thus, all results using pseudotyping systems must be confirmed using wild
type viruses (or chimeric CoVs, which better mimic wild type bat CoVs than pseudotyped viruses).
Finally, chimeric viruses that have been engineered to express “internal” (i.e. non-Spike) CoV proteins
have been used for testing the efficacy of MCMs targeting non-Spike proteins.1297 As with pseudotyped
viruses, due to significant differences in the course of infection between chimeric virus systems and wild
type viruses, such chimeric virus systems can be used to screen therapeutic candidates but do not replace
the need to test MCMs against the wild type virus.
The strengths and limitations of model systems that can be used for the development of broad-spectrum
CoV MCMs are summarized in Table 15.8. Taken together, chimeric bat-SARS CoV strains created using
GoF approaches that adapt a virus to a new host are uniquely capable of providing reliable information
about the broad-spectrum potential of CoV vaccines and therapeutics. Because most bat CoV strains
cannot be cultured, the use of wild type viruses cannot provide information about whether CoV MCMs
are capable of targeting a variety of SARS/MERS-like CoVs in addition to SARS and MERS. While
expressing CoV proteins in the context of other viruses (i.e. pseudotyped viruses and other chimeric virus
systems) may be useful for screening MCM candidates, all results must be confirmed using wild type
1295 Ibid.
1296 Ibid.
1297 Deng X et al (2014) A chimeric virus-mouse model system for evaluating the function and inhibition of papain-like
proteases of emerging coronaviruses. Journal of virology 88: 11825-11833

strains (or CoV chimeric strains) due to significant differences in the behavior of chimeric viruses versus
CoVs.

Table 15.8. Comparison of model systems for the development of broad-spectrum vaccines and therapeutics
• Enables testing of whether MCMs targeting any CoV
GoF: “Spike swapping” - chimeric CoVs protein or process confer broad-spectrum protection • Results using chimeric viruses may not
• Animal CoV plus SARS Spike [5] against multiple animal CoVs reflect the capacity of MCMs to target wild
• SARS plus animal CoV Spike [6] • Use of chimeric CoVs is more relevant to nature than type viruses
using mixed virus chimeras
Alt-GoF #1: Wild type animal CoVs • Use of wild type viruses is most relevant to nature • Most wild type animal CoVs cannot be
grown in culture
• Results may not be recapitulated in the

Alt-GoF #2: Pseudotyped viruses – express • Enables testing of whether MCMs targeting the Spike context of the wild type virus
CoV Spike proteins in the context of a protein confer broad-spectrum protection against multiple o Differential presentation of the Spike
different virus animal CoVs protein on the virus surface influences
antibody binding
• Results may not be recapitulated in the

Alt-GoF #3: Other mixed virus chimeras • Enables testing of whether MCMs targeting non-Spike context of the wild type virus
• Express “internal” (non-Spike) CoV proteins confer broad-spectrum protection against o Different course of infection and
proteins in other viruses multiple animal CoVs expression levels of CoV proteins affect
therapeutic efficacy

15.2 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances
Virus Production
15.2.1 Overview of the GoF Landscape: Approaches that Enhance the Production of Influenza
Viruses
This assessment describes the benefits of GoF experimental approaches that are reasonably anticipated to
enhance the production of influenza viruses. In this section, an overview of GoF approaches in this
phenotypic category is provided and the scientific outcomes and/or products of each approach are
described.
15.2.1.1 Generation of attenuated, high-yield candidate vaccine viruses through reassortment
Reassortment between a wild type strain and an attenuated, high-yield vaccine backbone strain generates
a “Candidate Vaccine Virus” (CVV), which comprises the HA and NA genes from the wild type strain
and the remaining six “internal genes” from the vaccine backbone strain. CVVs are attenuated and exhibit
higher levels of growth relative to the parental, wild type virus. CVVs may be generated through classical
reassortment methods, which involve co-infection of eggs or cells with the wild type strain and the
vaccine backbone strain followed by antibody-based selection for viruses with the correct surface
antigens, or through reverse genetics.1298 CVVs serve as the basis of vaccine strains that are used for the
production of influenza vaccines in eggs or cells. Additionally, in the context of academic research,
comparing the sequences of CVVs with varied growth properties enables the identification of mutations
that are associated with high yield.
15.2.1.2 Serial passaging of viruses in eggs or cells
Serial passaging of viruses in eggs or cells selects for higher-yield viruses. This approach is currently
used for the production of influenza vaccines in eggs or cells as well as for basic science research on the
mechanisms underlying high growth of influenza vaccine viruses. For vaccine production, manufacturers
serially passage CVVs in eggs or cells to generate high-yield vaccine seed strains that can be used for
large-scale production of vaccines. In the context of academic research, serial passaging of viruses in eggs
or cells followed by sequencing of the emergent higher-yield viruses enables the identification of
mutations that are sufficient to enhance the growth of the viruses. Subsequently, mutant viruses are
subjected to antigenic characterization using the hemagglutinin inhibition (HAI) assay or other assays to
identify which mutations confer high growth without changing the antigenicity of the strain. For research
purposes, this approach is most commonly carried out using vaccine backbone strains and CVVs but may
also be carried out using wild type strains.
15.2.1.3 Forward genetic screen to identify mutations that confer high growth to viruses
Forward genetic screens, which involve random mutagenesis of viruses followed by limited passaging to
select for mutants with high growth properties, enable the identification of mutations that confer high
growth to viruses. Forward genetic screens involving vaccine backbone strains and CVVs lead to the
identification of mutations that are sufficient to enhance the yields of vaccine viruses. Subsequently,
mutant viruses are subjected to antigenic characterization using the hemagglutinin inhibition (HAI) assay
backbone strain.

or other assays to determine which mutations confer high growth without altering the antigenicity of the
strain.
15.2.1.4 Targeted mutagenesis of viruses to introduce mutations that are associated with high growth
Targeted mutagenesis of viruses to introduce mutations that are associated with high growth, followed by
characterization of virus yields relative to the parental virus, demonstrates that a mutation or set of
mutations is necessary and sufficient to confer high growth. Subsequently, antigenic characterization
assays are performed to confirm that the mutations have not altered the antigenicity of the virus, and the
mutant strain is subjected to several rounds of passaging in eggs or cells to ensure that it is genetically
stable – that is, that it does not acquire additional mutations that alter its antigenicity upon further growth.
This knowledge provides a foundation for follow-up studies investigating the mechanistic basis of the
high-growth phenotype (e.g. the use of cell biological assays, biochemical assays, and other assays to
explore how the mutation enhances growth). Notably, these mutations may have been discovered through
a GoF approach, such as serial passaging or a forward genetic screen, or through an alt-GoF approach,
such as comparative analysis of wild type sequences.
Finally, it should be noted that experimental approaches involving targeted genetic modification of the
viral polymerase complex of avian viruses to render it more “human-like” (through site-directed
mutagenesis or reassortment between human and avian viruses) is also likely to enhance virus replication.
However, as the primary goal of those studies is to gain insight into the mechanisms underlying
adaptation of avian viruses to mammals, those studies are discussed in section 16.3 (“detailed analysis of
the benefits of GoF research that enhances mammalian adaptation and transmissibility”).
15.2.2 Overview of the potential benefits of GoF approaches that enhance the production of
influenza viruses
This section includes evaluation of whether GoF approaches that enhance virus production, described
above, have the potential to benefit each of the general benefit areas described in the NSABB’s
“Framework for Conducting Risk and Benefit Assessments of Gain-of-Function Research.” Each
potential benefit will be evaluated in detail below.
15.2.2.1 Scientific Knowledge Benefits
Information about genetic traits that confer high growth to vaccine viruses provides a foundation for
follow-up studies investigating the mechanistic basis of the enhanced growth phenotype, thereby
benefiting scientific knowledge about mechanisms underlying the high growth of vaccine viruses.
It should be noted that this type of GoF research has a clear translational focus, in that these studies aim to
learn how to modulate the phenotypic properties of attenuated, high-yield vaccine viruses rather than to
gain insight into the natural behavior of wildtype viruses.
All other GoF approaches are focused on identifying mutations that confer high growth to vaccine viruses
(either candidate vaccine viruses or vaccine backbone strains). Because these viruses have no correlate in
nature, this information does not inform the interpretation of genetic surveillance data from animals or
humans.

15.2.2.3 Development and Production of Vaccines
GoF approaches, namely the generation of attenuated, high-yield CVVs and serial passaging, are core
aspects of the existing processes for the production of influenza vaccines in eggs and cells, thus these
approaches currently benefit the production of influenza vaccines. The insights gleaned from GoF
approaches that enhance virus production also have the potential to improve vaccine production practices
in the future through two distinct mechanisms: (1) shortening vaccine production timelines, and (2)
improving the match between the virus strains used as the basis of vaccine strains and the strains that are
circulating during flu season (referred to as “vaccine match,” which is correlated with vaccine efficacy).
In brief, the former benefit derives from the creation of higher-yield vaccine viruses and the identification
of genetic traits that confer high growth to vaccine viruses, and the latter benefit derives from the creation
of genetically stable vaccine viruses that do not acquire antigenicity-altering mutations upon growth in
eggs or cells.
Information about mutations that confer high growth to vaccine viruses or about mutations that rescue the
growth of antiviral resistant strains is not relevant to the development of therapeutics.
inform diagnostic development.1299,1300
Since information about compensatory mutations that rescue the growth of antiviral resistant strains does
not inform assessments of the risk posed by circulating influenza strains, this information does not benefit
policy decisions about public health preparedness.
Similarly, information about mutations that confer high growth to vaccine viruses does not inform the
analysis of genetic surveillance data, so this information does not benefit policy decisions about public
health preparedness.
Increasing the yields of vaccine viruses, using information or products derived from GoF approaches that
enhance virus production, is likely to lower the cost per vaccine dose by enabling the production of a
greater number of vaccine doses using the same quantity of input materials. The economic benefits of
enhancements to vaccine virus yields were not described in detail in this report.
Because academic research investigating the mechanisms underlying high growth of vaccine viruses aims
to generate information or products that can be applied to vaccine production in order to address
shortcomings in the current process, first, an overview of existing systems for the production of influenza
vaccines is provided, including the role of GoF approaches. Then, the benefit of GoF approaches that are
currently used in influenza vaccine production for the availability and efficacy of influenza vaccines are
evaluated. In this sub-section, given the continued need for production of new seasonal influenza

vaccines, the potential for alternative approaches to provide the same or similar benefits in the immediate
future are also evaluated.
Next, shortcomings in the existing process for influenza vaccine production are reviewed; this motivates
the body of academic research that aims to improve the yields of vaccine viruses and the application of
that research to vaccine production. Then the potential for GoF approaches to identify genetic markers of
high growth and to advance foundational knowledge about mechanisms underlying high growth in ovo
and in cell culture, relative to alternative experimental approaches is evaluated. Finally, the potential for
the information/products derived from GoF research to further improve vaccine production practices,
relative to alternative experimental approaches and alternative scientific/technical innovations that can
similarly benefit the availability and efficacy of vaccines in the future, is evaluated.
15.2.3 Benefits of GoF Research that Enhances Production of Influenza Viruses to Current Vaccine
Production Practices
15.2.3.1 Current Processes for Production of Influenza Vaccines
To provide context for the evaluation of the benefits of GoF approaches to the current production of
influenza vaccines, first, a brief overview of existing influenza vaccine production processes is provided.
This review also provides important context for the subsequent discussion of the potential benefits of GoF
research to future vaccine production processes.
Because existing influenza vaccines rely predominantly on the immune response to the influenza HA
protein and are strain-specific, there is a continued need for production of new influenza vaccines to
protect public health. Specifically, seasonal influenza vaccines must be updated annually to accommodate
antigenic drift of circulating influenza viruses, and specific vaccines must be produced in response to the
emergence of a novel pandemic strain. Three different influenza vaccine production technologies have
been approved by the Food and Drug Administration (FDA): egg-based vaccines, cell-based vaccines,
and recombinant vaccines.1301 Egg- and cell-based vaccines are derived from whole viruses, whereas
recombinant vaccines are virus-free. The majority of influenza vaccines produced in the US are derived
from viruses grown in embryonated chicken eggs. Egg-grown viruses may be chemically inactivated and
delivered as a “flu shot,” a method of vaccine production that has been used for over 70 years, or
delivered as live attenuated vaccines in the form of a nasal spray.1302,1303,1304 Recently, a process using
cultured mammalian cells has been developed for the production of inactivated influenza vaccines; one
cell-based vaccine has been commercially available in the US since 2012.1305 Recombinant vaccines,
which are virus-free vaccines that are based on influenza proteins produced in insect cells or other protein
expression system, represent the newest production technology. One recombinant vaccine was FDA-
approved in 2013, and several others are in various stages of commercial development.1306,1307,1308
1301 How Influenza (Flu) Vaccines are Made. CDC. http://www.cdc.gov/flu/protect/vaccine/how-fluvaccine-made.htm. Last
Update Accessed September 14, 2015.
1302 Ibid.
1305 How Influenza (Flu) Vaccines are Made. CDC. http://www.cdc.gov/flu/protect/vaccine/how-fluvaccine-made.htm. Last
Update Accessed September 14, 2015.
1306 Ibid.
1308 Shaw A (2012) New technologies for new influenza vaccines. Vaccine 30: 4927-4933

The processes and timelines for production of egg- and cell-based vaccines, including inactivated and live
attenuated vaccines, are similar (Figure 15.1).

Figure 15.6. Timeline for egg- and cell-based production of influenza vaccines. Steps outlined in blue are managed by the WHO, steps outlined in black
are managed by the FDA, and steps outlined in pink are managed by vaccine manufacturers. Two steps – candidate vaccine virus (CVV) development
and vaccine seed strain development – involve GoF approaches and are highlighted in green text. The height of the bars reflects the yield of the virus at
that stage; trapezoidal stages indicate that virus yields are enhanced over the course of that step. The length of the bars reflects the average time needed
to complete that stage of the process. Clinical trials are not conducted for new seasonal influenza vaccines but are conducted for pandemic influenza
vaccines; the scale of clinical trials and length of this stage will vary depending on the strain. Overall, production of multivalent seasonal or monovalent
pandemic influenza vaccines spans six to eight months. For production of seasonal flu vaccines, this timeline begins with WHO strain selection (week 12
in the above timeline). For production of pandemic strains, this timeline begins with isolation of the field strain and CVV development (week 0 in the
above timeline).

First, a selected field isolate that is representative of circulating strains must be attenuated and its growth
in eggs/cells must be enhanced in order to be suitable for large-scale manufacturing of vaccine virus. This
growth enhancement is achieved through the use of two different GoF approaches. The first GoF
approach involves reassortment between a field isolate and an attenuated, high-yield “vaccine backbone
strain” to generate a CVV, as described above.1309 CVVs undergo a series of characterization assays
before they are released to manufacturers, including pathogenicity testing in ferrets, antigenic
characterization, and several rounds of passaging to ensure that mutations that lead to antigenic changes
will not arise during growth in eggs/cells.1310,1311,1312 Upon receipt of a CVV, vaccine manufacturers
serially passage the CVV in eggs or cells to increase its yield, representing the second GoF approach used
to enhance the yields of vaccine viruses during the vaccine production process.
Collectively, the result is a high-yield vaccine seed virus that can be used for large-scale production of
vaccine virus. In parallel to vaccine seed strain development, the FDA prepares “potency reagents” for the
single-radial immune-diffusion (SRID) assay used to standardize antigen quantities, namely HA antigen
and HA-specific antiserum produced in sheep.1313 Large-scale production of bulk antigen involves nested
cycles of virus production in eggs or cells, purification and processing of virus (including chemical
inactivation, if applicable), and quantification of HA antigen yields using the SRID assay. For production
of seasonal, multivalent vaccines, vaccine doses are formulated following consecutive production of
monovalent bulk antigen for each component of the vaccine (one A/H1N1 strain, one A/H3N2 strain, and
one or two B strains). 1314,1315,1316 New seasonal vaccines are not clinically tested each year, but pandemic
vaccines must undergo clinical trials to establish the safety of the vaccine and determine the dosing
parameters needed to elicit a strong immune response (e.g. amount of antigen, number of doses, etc.).
Manufacturers set aside an initial lot(s) of vaccine antigen for clinical trial use, and the trials are
conducted in parallel with additional bulk antigen production.1317,1318 Finally, all lots of seasonal and
pandemic vaccines are safety-tested and FDA-approved prior to release.
Overall, the production of egg- and cell-based influenza vaccines requires six to eight months.1319,1320 For
production of pandemic vaccines, this timeline begins with the selection of a field isolate to be used as the
1309 It should be noted that although CVVs are usually 6:2 reassortants (i.e. comprising the HA and NA genes from the field
isolate and all other genes from the vaccine backbone strain), CVVs may also be 5:3 reassortants (e.g. HA, NA, and other
gene from the field isolate, and the remaining five genes from the vaccine backbone strain).
1310 Vaccine response to the avian influenza A(H7N9) outbreak- step 1: development and distribution of candidate vaccine
viruses. http://www.who.int/influenza/vaccines/virus/CandidateVaccineVirusesH7N9_02May13.pdf. Last Update Accessed
September 14, 2015.
1311 Update of WHO biosafety risk assessment and guidelines for the production and quality control of human influenza
vaccines against avian influenza A(H7N9) virus.
http://www.who.int/biologicals/areas/vaccines/influenza/biosafety_risk_assessment_10may2013.pdf. Last Update Accessed
September 14, 2015.
1312 (2015q) Current practices in influenza vaccine production. Interview with Industry or Federal Government Representative
1314 Food and Drug Administration. Annex 5: Vaccination Development and Production - Draft
http://www.hsdl.org/?view&did=459937. Last Update Accessed September 15, 2015.
1317 Ibid.
1318 Cho D. Regulatory Pathways for Registration of Seasonal and Pandemic Influenza Vaccines: FDA Approach.
http://www.who.int/phi/Day2_2_Cho_FDA_approach_Flu_vax_PM_Dubai2013.pdf. Last Update 19 March 2013. Accessed
14 September 2015

basis of the vaccine strain and includes the time needed for development and testing of the CVV. For
seasonal influenza vaccines, production is initiated following strain selection by the WHO in February or
September (for the Northern and Southern hemispheres, respectively). The WHO Global Influenza
Surveillance and Response System (GISRS) oversees the development and testing of new CVVs
throughout the year, when antigenically distinct strains emerge, and the strain selection committee
recommends strains for which antigenically similar CVVs are available.1321
15.2.3.2 GoF Approaches Needed to Maintain Current Influenza Vaccine Production Systems
Because the strain composition of influenza vaccines must be updated annually, the CDC’s Advisory
Committee on Immunization Practices recommends annual influenza vaccination for all people ages six
months and older.1322 Currently, over 99% of influenza vaccines used in the US are produced in eggs or
cells,1323,1324 which relies on GoF approaches for two stages of the production process: CVV development
and vaccine seed strain production (Figure 15.1). As described above, each of those GoF approaches
enhances virus production, collectively increasing HA antigen yield at least 12-fold relative to the cognate
wildtype strain. 1325 Altogether, these approaches, which are used throughout the egg- and cell-based
vaccine manufacturing industry, result in the production of over 170 million doses of seasonal influenza
vaccine annually.1326 It should also be noted that attenuated, high-yield candidate vaccine viruses have
been used for the production of influenza vaccines since 1971.1327,1328,1329
Because of the continued need for production of seasonal influenza vaccines, as well as the need to
maintain robust capabilities for the production of pandemic vaccines for pandemic preparedness,
alternative approaches must similarly benefit vaccine production in the immediate future. Eliminating
GoF approaches from existing production processes would necessitate the use of vaccine viruses with
wild type growth properties, which could be achieved through the direct use of field isolates or through
the use of novel reassortants that are attenuated but do not exhibit enhanced yields (Table 15.9).
1321 (WHO) WHO. Recommended composition of influenza virus vaccines for use in the 2015- 2016 northern hemisphere
influenza season. http://www.who.int/influenza/vaccines/virus/recommendations/201502_recommendation.pdf?ua=1. Last
Update February 26, 2015. Accessed October 20, 2015.
1322 CDC’s Advisory Committee on Immunization Practices (ACIP) Recommends Universal Annual Influenza Vaccination.
http://www.cdc.gov/media/pressrel/2010/r100224.htm. Last Update Accessed September 15, 2015.
1327 Kilbourne ED (2006) Influenza pandemics of the 20th century. Emerging infectious diseases 12: 9-14
1328 World Health Organization. Influenza vaccine viruses and reagents. http://www.who.int/influenza/vaccines/virus/en/. Last
Update September 2015. Accessed 30 September 2015.
1329 Nesterova D. Influenza Vaccine History. http://www.vaccination.english.vt.edu/wp-content/uploads/2015/04/updated-
influenza-media-kit-4.pdf. Last Update October 2012. Accessed 30 September 2015.

Table 15.9. Summary of the Benefits of GoF Approaches that Enhance Virus Production
Vaccine Development Benefits – Current Influenza Vaccine Production Practices
Experimental Approach Benefits Limitations Barriers
• Annual production of > 170
GoF: Use of high-growth reassortant million doses of seasonal
strains for vaccine production (status influenza vaccine
quo), which exhibit:
• Ability to release pandemic flu N/A (discussed elsewhere) None (current system)
• Enhanced virus production vaccine ~ 8 months after
• Attenuated virulence emergence of a novel pandemic
strain
• Adverse consequences for • Construction of new

Alt-GoF #1: Use of wild type strains • Avoid use of vaccine strains vaccine availability manufacturing facilities capable
for vaccine production with enhanced yield relative to o Inability to produce vaccine of large-scale production of wild
wild type viruses that meets FDA purity type viruses that are pathogenic
standards to humans
o Significantly reduced rates
of vaccine production
• Adverse consequences for • Requires development of new
Alt-GoF #2: Use of novel reassortant vaccine match vaccine backbone strains that are
strains that are: o Prioritize growth properties attenuated but do not confer
• Avoid use of vaccine strains high growth
• Attenuated with enhanced yield relative to over antigenic properties
when choosing strains for o Commercial use of new
• Exhibit wild type levels of virus wild type viruses
vaccine production vaccine backbone strains
production may require FDA approval
o Choose seasonal strains for
vaccine at least one year in
advance of the start of the
target flu season

Vaccine Development Benefits – Current Influenza Vaccine Production Practices
Experimental Approach Benefits Limitations Barriers
• Only one recombinant flu
vaccine is currently FDA-
Alt-GoF #3: Use of alternative, virus- • Avoid use of vaccine strains
approved (Flublok) • Development and registration of
free vaccine platforms with enhanced yield relative to new influenza vaccines is a
wild type viruses o Use limited to people 18
lengthy and expensive process
• Recombinant vaccines, DNA- years and older
(8 – 10 years and 0.3 – 1 billion
based vaccines • Additional benefits discussed
o Represented less than 0.1% dollars)
further below
of vaccine distributed during
2014 – 2015 flu season

Using viruses with wild type growth properties in lieu of high-yield vaccine viruses generated using GoF
approaches would have adverse consequences for vaccine availability. Figure 16.2 illustrates three
different scenarios associated with the use of wild type viruses for egg- and cell-based production of
influenza vaccines, which would impact several stages of the vaccine production process. Specifically,
use of wild type viruses in lieu of high-yield vaccine viruses would:1330
• Eliminate the need for CVV development and CVV testing, shortening the vaccine production
timeline by approximately 9 weeks,
• But would reduce the rate of bulk antigen production (i.e. by 12-fold, on average), and
• Minimally affect the time needed for seed strain development,1331 potency reagent development,
vaccine formulation, or lot testing/release.
Most influenza viruses grow poorly in eggs and cells. If manufacturers attempted to use strains with poor
growth properties for large-scale infection of eggs/cells, the quantity of virus produced would likely be
low enough, relative to egg/cellular proteins, that existing manufacturing processes would fail to produce
“purified” antigen that meets FDA purity standards. This manufacturing failure would result in no
vaccine produced (Figure 15.2, scenario 3). At best, manufacturers could pause production and attempt
to adjust their purification protocols, which would extend an already lengthy production process.1332
Alternatively, a field isolate with exceptional growth properties that permits production of bulk antigen at
reduced rates could be used to produce the same number of doses currently produced over an extended
period of time (Figure 15.2, scenario 1) or to produce a smaller number of doses on the current production
timescale (Figure 15.2, scenario 2). (It should be noted that influenza vaccine production experts deemed
this scenario – field isolates with unusually high yields and correct antigenic properties – highly
unlikely.)1333 To illustrate the consequences for vaccine availability in scenarios 1 and 2, there is an
assumption that the yields of the exceptional field isolate are approximately one-third those of a typical
seasonal H1N1 strain.1334,1335,1336 Use of such an isolate to produce the same number of doses would
lengthen the time needed for bulk antigen production by three-fold, from 19 weeks to 57 weeks, which,
coupled with six weeks for seed strain development, would result in release of vaccine 63 weeks after
initiation of manufacturing. During the 2009 H1N1 pandemic, this vaccine would not have been available
until April 2015, well after peak waves of flu activity and near the end of the pandemic
1331 Industry representatives noted that whether a high-growth reassortant or a field isolate were used for large-scale production,
some degree of passaging by manufacturers is required for optimizing infection conditions using the particular strain and for
preparing enough seed virus for large-scale infection of eggs/cells.

1333 Ibid.
1334 During the 2009 H1N1 influenza pandemic, the yields of the initial H1N1 pdm high-growth reassortant (HGR) strain were
approximately one-third those of a typical seasonal H1N1 HGR. This strain was used to produce clinical lot material, thus
demonstrating that this yield reduction does not preclude preparation of sufficiently pure antigen. However, subsequently,
the initial HGR was extensively passaged to increase its yield to enable preparation of sufficient quantities of vaccine in a
timely manner.

1336 WHO. Availability of a new candidate reassortant vaccine virus for pandemic (H1N1) 2009 virus vaccine
development http://www.who.int/csr/disease/swineflu/guidance/vaccines/candidates/cp122_2009_0608_availability_of_new
_cr_vaccine_virus_nibrg-121-final.pdf?ua=1. Last Update Accessed September 15, 2015.
GoF RBA Report Gryphon
Scientific, LLC
575
period.1337,1338,1339,1340 In the context of seasonal influenza vaccine production, this production timeline
would necessitate strain selection more than one year in advance of the start of the target flu season.
Given the challenges for such long-term predictions of the dominant circulating strains and the likelihood
of antigenic drift over the course of the production year, the vaccine strains would be highly unlikely to
match the circulating strains during the target flu season, leading to reduced vaccine efficacy.1341
Alternatively, the exceptional field isolate could be used to produce a smaller number of doses on the
standard production timescale (scenario 2). During a pandemic, this shortcoming would translate to a
two-fold reduction in vaccine availability, while use of such a field isolate for seasonal flu vaccine
production would result in production of one-third the typical number of doses (i.e. enough doses to
vaccinate just under 20% of the US population).1342,1343 Furthermore, in either scenario, the choice of a
vaccine strain would be guided by the growth properties of strains of interest, which may lead to the
production of vaccines that poorly match the antigenicity of the dominant circulating strain.1344 Finally, it
is noted that use of an attenuated field isolate would add nine weeks to the timelines described above, for
development and testing of the attenuated reassortant, further delaying release of the vaccine and/or
reducing the number of doses produced.
1337 The US Public Health Emergency for H1N1 influenza expired on June 23, 2010, and the CDC’s official estimates for
pandemic H1N1-associated morbidity and mortality in the US span April, 12 2009 through April 10, 2010.
1338 CDC. 2009 H1N1 Flu. http://www.cdc.gov/h1n1flu/. Last Update Accessed September 15, 2015.
1339 Shrestha SS et al (2011) Estimating the burden of 2009 pandemic influenza A (H1N1) in the United States (April 2009-
April 2010). Clinical infectious diseases : an official publication of the Infectious Diseases Society of America 52 Suppl 1:
S75-82

1341 (2015e) Influenza Vaccine Strain Selection. Interview with Academic Researcher or Federal Government Representative
1342 Current vaccine production processes lead to vaccine release at approximately week 37 following emergence of a new
pandemic strain. Using a field isolate, this timeline would comprise strain isolation (2 weeks), vaccine seed strain
development (6 weeks), and large-scale vaccine production (29 weeks). Given production at one-third of the typical rate,
this would result in approximately half of the number of doses produced relative to use of a standard strain over a 19-week
production period.
1343 Because CVVs for seasonal influenza strains are produced in advance of strain selection, using field isolates in lieu of CVVs
would not alter the basic components of the industrial production process. Thus, use of a field isolate with virus yields
approximately one-third those of a typical high-growth reassortant would lead to the production of approximately one-third
the typical amount of vaccine over the course of the same time period. As approximately 170 million doses of influenza
vaccine are produced annually, this would result in production of about 55 million doses, or enough to vaccinate 18% of the
US population.

GoF RBA Report Gryphon
Scientific, LLC
576
Figure 15.2. Consequences for influenza vaccine production timelines if strains with field-like growth properties were used in lieu of high-growth
reassortants generated through GoF approaches. In Scenario 3, the growth properties of the field strain are so low that virus antigen cannot be purified
to FDA standards; thus, no vaccine is produced. In Scenarios 1 and 2, it is assumed that a field isolate with exceptional growth properties (4-fold greater
than average, leading to production of bulk antigen at approximately one-third the average rate), is used. This strain could be used to produce the same
number of doses over a greater period of time (more than one year, Scenario 1) or could be used to produce a smaller number of doses in the same
period of time (two- to three-fold fewer doses, Scenario 2). In either Scenarios 1 or 2, manufacturers are likely to prioritize growth properties of the
strain over antigenicity, leading to potentially poor vaccine match and reduced vaccine efficacy.

Additionally, neither alternative (i.e. use of wild type strains or use of novel reassortants with wild type
growth properties) can be implemented immediately. Large-scale production of field isolates for the
purpose of producing inactivated vaccines would pose significant risks to vaccine manufacturers prior to
the inactivation step, presumably requiring the construction of new manufacturing facilities capable of
virus production under higher biocontainment conditions. Of note, field isolates cannot be used as a basis
for live vaccines due to their pathogenicity. The alternative, use of attenuated vaccine viruses with wild
type growth properties, would necessitate the development, and perhaps subsequent FDA licensing, of
novel vaccine backbone strains that attenuate but do not confer high growth to reassortant viruses.
As described above, production of virus-based vaccines in eggs/cells necessitates passaging of the

antigenic strain of interest to produce enough stock virus to infect eggs/cells for large-scale
manufacturing, which inevitably selects for higher-yield viruses due to the high mutation rate of influenza
viruses.1345 If this passaging were considered to be a GoF approach, in addition to the approaches
described above that deliberately enhance the yields of vaccine viruses, then completely avoiding
manipulations that are reasonably expected to enhance virus production precludes production of egg- and
cell-based influenza vaccines. In that case, virus-free vaccine platforms, such as recombinant or DNA-
based vaccines, represent an alternative to egg- and cell-based flu vaccines (Table 15.9).1346,1347,1348
However, the one recombinant flu vaccine that is commercially available is only approved for use in
people 18 years of age and older and represented just 50,000 of more than 140 million doses administered
during the 2014 – 2015 flu season.1349,1350 Although other recombinant vaccines are in late stages of
development, given the long and expensive product development cycle for new influenza vaccines –
spanning 8 – 12 years and costing 300 million to one billion dollars including research, clinical
development, and registration with the FDA – alternative, virus-free flu vaccine platforms are not a viable
replacement for egg- and cell-based vaccines in the immediate future.1351
15.2.3.3 Summary – Benefits of GoF approaches relative to alt-GoF approaches for current influenza
vaccine production
The strengths and limitations of GoF and alt-GoF approaches that could be used for the current
production of influenza vaccines are summarized in Table 15.9. Taken together, this analysis
demonstrates that GoF approaches to enhance the growth of attenuated vaccine strains are a uniquely
critical component of the current ability to produce sufficient and effective vaccines for seasonal and
pandemic influenza. The use of field strains or of novel reassortant strains with field-like growth
properties for egg- and cell-based vaccine production would have adverse consequences for the
availability and efficacy of vaccines, including the possibility that no vaccine could be produced, and
neither approach could be implemented immediately. Recombinant vaccines and other virus-free vaccine
platforms represent a promising approach for future influenza vaccine production, but the one
recombinant vaccine that is currently licensed represents less than 1% of seasonal influenza vaccines
administered annually, and lengthy regulatory processes will delay the availability of additional virus-free
vaccines in the future.
1347 Kim JH, Jacob J (2009) DNA vaccines against influenza viruses. Current topics in microbiology and immunology 333: 197-
210
1350 Protein Sciences. Flublok. http://www.proteinsciences.com/FVAC.htm. Last Update Accessed September 15, 2015.

15.2.4 Benefits of GoF Research that Enhances Production of Influenza Viruses to Scientific
Knowledge and to Future Influenza Vaccine Production Practices
In this section of the report, the benefits of GoF research that enhances virus production to scientific
knowledge and to vaccine production in the future are evaluated. As noted above, academic research in
this phenotypic category is focused on enhancing the yields of vaccine viruses and has a clear
translational focus, on generating higher-yield vaccine strains that can be used for vaccine production,
generating information about high-yield markers that can be incorporated into vaccine strains, and/or
deepening understanding of the mechanisms regulating the growth of vaccine viruses to provide a
foundation for the development of higher-yield vaccine strains in the future. To provide context for these
experimental goals, first, shortcomings in the current system for production of influenza vaccines are
reviewed. Next, the potential benefits of GoF research to scientific knowledge about the genetic and
phenotypic traits underlying high-growth of influenza viruses in eggs and cells, relative to alternative
experimental approaches are evaluated. Finally, the section concludes with evaluation of how the insights
and products arising from GoF research may be applied to vaccine production to further improve existing
production practices and benefit public health in the future.
15.2.4.1 Shortcomings of Current Systems for Production of Influenza Vaccines
Interviews with stakeholders in the influenza research and public health communities highlighted that the
lengthy production timelines for existing egg- and cell-based vaccines critically limit the mitigating
impact of influenza vaccination on the morbidity and mortality associated with influenza outbreaks. In the
context of seasonal flu epidemics, existing production timelines necessitate strain selection nine months in
advance of the peak of the target flu season.1352 As a result, one or more vaccine strains are often
imperfectly matched to circulating strains, either due to poor strain selection (i.e. incorrect prediction of
which strain would predominate in nature) or antigenic drift of the selected strain in nature during the
course of vaccine production, which reduces the efficacy of the vaccine.1353 In the context of pandemics,
vaccines are simply unavailable to protect the public until at least six months into the outbreak.1354,1355
Additionally, CVVs may acquire mutations that alter their antigenicity during growth in eggs or in cells, a
third shortcoming that results in poor vaccine match and that can affect the production of seasonal and
pandemic vaccines. In particular, H3N2 strains often acquire antigenicity-altering mutations upon growth
in eggs, which is especially concerning given that H3N2 strains tend to cause more severe disease than
H1N1 strains.1356,1357,1358,1359
The yields of vaccine viruses establish the rate of bulk antigen production and thus serve as a key
determinant of the time needed for vaccine production. Some strains, including H3N2 strains and many
1352 Ibid.
1356 (2015y) Candidate vaccine virus development. Interviews with Influenza Researchers Involved in Candidate Vaccine Virus
Development.
1358 Huang SSH et al (2011) Comparative Analyses of Pandemic H1N1 and Seasonal H1N1, H3N2, and Influenza B Infections
Depict Distinct Clinical Pictures in Ferrets. PLoS ONE 6: e27512
1359 Kaji M et al (2003) Differences in clinical features between influenza A H1N1, A H3N2, and B in adult patients.
Respirology (Carlton, Vic) 8: 231-233

zoonotic influenza strains, routinely produce low-yield CVVs, and any strain may unexpectedly produce a
poorly growing CVV.1360,1361 For example, the 2009 H1N1 pandemic CVV exhibited production yields
approximately one-third those of a typical H1N1 seasonal CVV.1362 In either case, the need to extensively
passage a low-yield CVV to render it suitable for large-scale production, as happened in 2009, and/or to
utilize a sub-par CVV for production, delays manufacturing and subsequent release of the vaccine.1363,1364
Additionally, even high-growth CVVs typically exhibit reduced yields relative to vaccine backbone
strains, indicating that CVV yields could be further increased. Thus, the limited production yields of
CVVs represent a gap that compromises the efficacy and utility of existing influenza vaccines by
lengthening egg- and cell-based vaccine production timelines. Furthermore, the fact that existing
strategies for CVV development do not consistently produce high-yield strains highlights the incomplete
understanding of the genetic determinants underlying high growth in eggs and cells.
15.2.4.2 Benefits of GoF Research that Enhances Virus Production to Scientific Knowledge
Several GoF approaches can be used to discover mutations associated with high growth of vaccine
backbone strains and CVVs. Serial passaging of viruses in eggs or cells is a classic method for identifying
mutations that confer enhanced growth, while forward genetic screens, which involve randomly
mutagenizing strains and subsequent passaging of mutant libraries to select for high-growth variants,
represent a modern approach for discovery of genetic markers associated with high growth. Both
approaches enable the discovery of mutations that are sufficient to confer higher-than-wild type levels of
growth to any virus strain of interest. However, both approaches are limited by their narrow breadth; that
is, the mutations that are identified may confer high growth to the studied strain only.
Comparing the sequences of CVVs with varied growth properties is another GoF method that can be used
to identify mutations that are associated with high growth. (It should be noted that this method is
considered a GoF approach because CVVs exhibit enhanced replication relative to vaccine backbone
strains, as described above.) However, unlike serial passaging and forward genetics approaches,
comparative sequence analysis is unlikely to uncover genetic markers associated with greater-than-wild
type levels of growth because it is limited to analysis of existing isolates.
In either case, the phenotypic consequences of mutations can then be confirmed through targeted
mutagenesis of the parental strain. Collectively, these approaches enable the identification of genetic traits
that are necessary and sufficient to confer higher-than-wild type levels of growth to vaccine viruses, for
any strain of interest. This information provides a foundation for follow-up structural, biochemical, and
cell biological assays investigating the phenotypic consequences of the mutations, in order to gain insight
into the mechanisms underlying the enhanced growth phenotype. Subsequently, using targeted
mutagenesis to determine the effect of the marker on virus growth in a new strain context provides insight
into whether the marker is likely to be broadly useful for improving CVV yields as well as whether the
phenotypic traits underlying high growth are conserved across strains.
Development.
1363 Ibid.
1364 WHO. Availability of a new candidate reassortant vaccine virus for pandemic (H1N1) 2009
virus vaccine
development http://www.who.int/csr/disease/swineflu/guidance/vaccines/candidates/cp122_2009_0608_availability_of_new
_cr_vaccine_virus_nibrg-121-final.pdf?ua=1. Last Update Accessed September 15, 2015.

15.2.4.2.2 Benefits and limitations of alt-GoF approaches
Alternative experimental approaches (“alt-GoF”) can also be used to uncover genetic markers associated
with high growth. Sequence comparison of wildtype strains with varied growth properties may provide
insight into mutations that confer a growth advantage. Of note, because of the importance of genetic
context on multi-genic traits such as fitness, mutations that confer high growth to wildtype strains may
not confer high growth to vaccine strains (i.e. reassortants that include the HA and NA from the field
isolate and the remaining six genes from a vaccine backbone strain). Similar to comparative sequence
analysis of CVVs, this approach depends on the existence of high-growth strains in nature and cannot
identify mutations that confer exceptional yields.
Genetic screens to identify mutations that reduce growth (i.e. loss-of-function, or LoF) can lead to the
discovery of mutations that are necessary for growth. A major limitation of this approach is that it may
uncover mutations that reduce growth for “trivial,” reasons, i.e. that modulate critical aspects of virus
function that are necessary for viability but do not directly contribute to high growth. An additional
drawback is that it is much less efficient than its GoF counterpart because mutants must be screened for
reduced growth (versus selection for high growth through passaging). Finally, the utility of the
information gleaned from LoF screens also depends on the existence of high-growth strains in nature.
LoF approaches may also be used to confirm that a particular amino acid residue (discovered through
GoF or alt-GoF approaches) is necessary for high growth. However, the marker may not be sufficient to
enhance growth if introduced into a different strain, limiting the utility of this result for vaccine
production.
15.2.4.2.3 Summary – Benefits of GoF approaches relative to alt-GoF approaches for scientific
knowledge
section, with respect to the ability of each approach to provide insight into the mechanisms underlying the
growth of influenza vaccine viruses, are summarized in Table 15.10. GoF approaches are uniquely
capable of discovering mutations that enhance the growth of any vaccine virus strain to greater-than-
wildtype levels. In addition, GoF approaches are uniquely capable of demonstrating that particular
mutations are necessary and sufficient to enhance the growth of vaccine viruses. Together, this
information provides a strong foundation for follow-up studies investigating the mechanistic basis of high
growth of vaccine viruses.
Alternative approaches have significant limitations for the study of mechanisms governing the growth of
vaccine viruses. Comparative sequence analysis of wild type isolates is limited to the study of phenotypes
underlying naturally high levels of growth, and the information gleaned from these studies may not
translate to vaccine viruses. LoF approaches are inefficient, and genetic markers that are necessary for
high growth may not be sufficient to enhance growth if introduced into a different strain.
Furthermore, GoF approaches to confirm that particular markers confer high growth are uniquely critical
for generating information that can be translated to the vaccine production process. The phenotypic
consequences of incorporating mutations that are associated with high growth or that are necessary for
high growth into vaccine viruses are too uncertain to be applied to vaccine production.

Scientific Knowledge Benefits – What is the mechanistic basis of high growth of influenza viruses in eggs and cells?
Experimental Approach Benefits Limitations

GoF #1 [4]*: Serial passaging of viruses in eggs • Identify new genetic traits
or cells
• Associative – whether mutations are necessary to enhance growth must
that are sufficient to be experimentally confirmed
GoF #2 [6]: Forward genetic screen to identify enhance the growth of any
virus to greater than wild • Narrow breadth – results may not generalize to other influenza strains
mutations sufficient to confer increased virus
production on virus backbones. type levels
• Identify genetic traits that • Associative - whether mutations are necessary and sufficient to
GoF #3 [7]: Comparative sequence analysis of enhance growth must be experimentally confirmed
CVVs with varied growth properties to identify are associated with
genetic traits associated with high growth naturally high levels of • Utility depends on the availability of CVVs with varied growth
growth of existing CVVs properties
GoF #4 [8,9]: Targeted genetic modification of
parental virus to introduce mutations shown to • Identify genetic traits that
be associated with enhanced growth are necessary and
• Narrow breadth - results may not generalize to other influenza strains
• Confirm the phenotypic effects of a particular sufficient to enhance the
mutation in a known strain or validate its growth of any virus
phenotypic effects in a new strain context
• Associative - whether mutations are necessary and sufficient to

Alt-GoF #1 [1]: Comparative sequence analysis • Identify genetic traits that enhance growth must be experimentally confirmed
are associated with
of wild type strains with varied growth
naturally high levels of
• Utility depends on the availability of field isolates with varied growth
properties to identify genetic traits associated properties
with high growth growth of existing field
strains • Epistasis – mutations that confer high growth to wild type strains may
not be conserved in vaccine strains
Alt-GoF #2 [2,3]: Loss of Function approaches
• Genetic markers may not be sufficient to enhance growth in a different
• Forward genetic screen to identify new • Identify genetic traits that strain context
mutations that attenuate virus production are necessary for naturally
high growth of existing
• Inefficient – screening for attenuated growth is less efficient than
• Targeted genetic modification of parental selecting for enhanced growth
virus to mutate amino acid residues associated field strains
with high growth • Narrow breadth - results may not generalize to other influenza strains

Scientific Knowledge Benefits – What is the mechanistic basis of high growth of influenza viruses in eggs and cells?


15.2.4.3 Public Health Benefits of GoF Research that Enhances Virus Production
GoF approaches that enhance virus production have the potential to improve existing vaccine production
practices by addressing two shortcomings in the current process for egg- and cell-based vaccine
production: (1) some strains acquire mutations that alter antigenicity during growth in eggs or cells,
leading to poor vaccine match, and (2) production timelines are too long. As described above, the yield of
CVVs governs the rate of bulk antigen production in eggs/cells and thus serves as a key determinant of
the length of time needed for egg- and cell-based vaccine production. Lengthy vaccine production
timelines impact the quality and availability of seasonal and pandemic flu vaccines differently. In the
context of seasonal flu epidemics, existing production timelines necessitate strain selection nine months in
advance of the peak of the target flu season.1365 As a result, one or more vaccine strains are often
imperfectly matched to circulating strains, which reduces the efficacy of the vaccine.1366 In the context of
pandemics, vaccines are simply unavailable to protect the public until at least six months into the
outbreak.1367 (It is noted that the impact of improving vaccine availability and efficacy during influenza
pandemics and seasonal epidemics will be further explored using quantitative methods in the quantitative
benefit assessment section of this report (section 9.12).)
This sub-section first evaluates how GoF research may benefit the availability and efficacy of vaccines by
generating genetically stable, high-yield CVVs. Then, alternative approaches that have potential to
similarly benefit vaccine production by shortening vaccine production timelines are evaluated. Finally,
alternative scientific and technical innovations that may improve the quality and availability of vaccines
through completely different mechanisms are analyzed.
15.2.4.3.1 Benefits of GoF Approaches to Future Influenza Vaccine Production
GoF research that generates genetically stable, higher-yield CVVs can be translated to vaccine production
through direct use of lab-generated CVVs or through incorporation of genetic markers that confer high-
growth into existing CVVs using targeted mutagenesis. Of note, studies that increase the yields of vaccine
backbone viruses generate more broadly applicable information than those focusing on particular CVVs.
As described above, this research can address two shortcomings in the current vaccine production
process. First, information about genetic markers that confer high growth without altering antigenicity can
benefit the production of vaccines for strains that readily mutate during passage in eggs or cells, such as
H3N2 strains. Specifically, the use of new, GoF-derived genetically stable CVVs would enable the
production of vaccines that match the antigenicity of the selected strains, which translates to improved
vaccine efficacy. Second, the use of higher-yield vaccine viruses or the incorporation of high-growth
markers into existing CVVs can benefit the production of vaccines for any strain by increasing the rate of
bulk antigen production and thereby shortening vaccine production timelines.
One key constraint on the benefits afforded by improvements to CVV yields is the limited production
capacity of eggs and cells. Current egg-based vaccine production systems are at or near maximal levels of
production, suggesting that the benefits of GoF research are largely limited to improving the growth of
“poor” CVVs.1368 However, because many CVVs based on zoonotic viruses and seasonal H3N2 viruses

grow poorly in eggs, simply improving their production would significantly benefit public health.1369,1370
In contrast, the production capacities of cell-based systems have not yet plateaued, thus GoF research that
improves CVV yields has the potential to benefit production of vaccines for all influenza sub-types using
cell-based systems.1371
Importantly, because these minor modifications to existing CVVs are not likely require FDA approval for
use in vaccine production, these benefits can be realized in the immediate future.1372
15.2.4.3.2 Benefits of Alternative Approaches with Potential to Shorten Vaccine Production Timelines
Several alternative approaches have potential to improve the availability and efficacy of vaccines by
shortening vaccine production timelines through different mechanisms. First, an alternative approach for
improving vaccine virus yields without enhancing the inherent growth properties of CVVs is through
modulation of the host cells that are used to produce virus. Specifically, identification of host genes that
suppress viral growth provides a basis for development of specialized knockout cell lines that permit
higher virus yields.1373 The key drawbacks to this approach are that research on whether such cell lines
will support high growth of a wide variety of influenza strains is limited, and currently, only one cell-
based vaccine that could potentially make use of this technology is licensed in the US.1374 Furthermore,
cell lines must undergo extensive testing in order to be FDA-approved for influenza vaccine production
prior to their commercial use, which will delay realization of this benefit.1375,1376 Finally, the risk
associated with GoF experiments that enhance virus production inheres in the fact that researchers are
working with increased viral titers relative to experiments using wildtype strains should be noted. As the
host cell modulation approach leads to the same consequence – i.e. that researchers handle higher
quantities of virus – this alt-GoF approach does not reduce risk relative to GoF approaches that enhance
viral titer through modulation of the virus.
An adjuvant is a substance that is added to a vaccine to boost the body’s immune response to the vaccine,
and including an adjuvant in a vaccine may enable the use of a smaller quantity of antigen to induce the
same level of protection (“dose sparing”).1377 Thus, incorporating adjuvants into existing egg- and cell-
based vaccines represents a different strategy for shortening production timelines, by enabling production
of the same number of doses over a shorter period of time. Most licensed vaccines in the U.S. are not
adjuvanted – one seasonal vaccine containing adjuvants was recently approved for use in people aged 65
Development.
1372 Ibid.
1373 Hamamoto I et al (2013) High yield production of influenza virus in Madin Darby canine kidney (MDCK) cells with stable
knockdown of IRF7. PloS one 8: e59892
1376 FDA. Guidance for Industry: Characterization and Qualification of Cell Substrates and Other Biological Materials Used in
the Production of Viral Vaccines for Infectious Disease Indications http://www.fda.gov/downloads/biologicsbloodvaccines/
guidancecomplianceregulatoryinformation/guidances/
vaccines/ucm202439.pdf. Last Update Accessed September 15, 2015.
1377 CDC. Vaccine Adjuvants. http://www.cdc.gov/vaccinesafety/concerns/adjuvants.html. Last Update Accessed September 15,
2015.

and older, and one licensed pandemic influenza vaccine contains adjuvants.1378,1379,1380,1381 Nonetheless,
use of adjuvants to improve the immunogenicity of seasonal influenza vaccines is an active area of
research. The major barrier to realization of this benefit is that existing vaccines that are re-formulated
with adjuvant are considered new drugs by the FDA and as such must undergo the standard licensure
pathway for unadjuvanted vaccines.1382,1383,1384 Although new seasonal inactivated influenza vaccines may
be considered for the accelerated regulatory pathway, which requires less extensive clinical trials than the
traditional regulatory pathway (coupled with industry commitment to post-licensure studies), even the
accelerated pathway spans over five years.1385,1386
Developing new vaccine platforms with faster production timelines represents a third alternative approach
for shortening the time needed for production of strain-specific vaccines. Recombinant vaccines, which
are virus-free vaccines comprised of recombinant influenza proteins produced in insect cells or other
protein expression systems such as plants, represent the most developed and promising approach.1387,1388
The major benefit of recombinant vaccines is that production can be rapidly scaled up in response to the
emergence of a novel pandemic strain, leading to production of clinical trial material one to two months
sooner than egg- and cell-based production systems and commercial release of vaccine six to eight weeks
sooner than traditional platforms. 1389 Although only one recombinant vaccine is currently FDA-licensed,
several other recombinant vaccines are in late stages of development, and experts in the influenza vaccine
field expect the production and use of this type of vaccine to increase over the next several
decades.1390,1391 However, as mentioned above, the time needed for completion of clinical trials and
1378 Ibid.
1379 Influenza A (H5N1) Virus Monovalent Vaccine, Adjuvanted.
http://www.fda.gov/BiologicsBloodVaccines/Vaccines/ApprovedProducts/ucm376289.htm. Last Update Accessed
September 15, 2015.
1380 FDA. FDA approves first seasonal influenza vaccine containing an adjuvant. FDA News Release.
http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm474295.htm. Last Update November 24, 2015.
1381
Novartis. FLUAD® (MF59®-Adjuvanted Influenza Vaccine) Fact Sheet.
https://www.novartis.com/sites/www.novartis.com/files/Fluad_Fact_Sheet.pdf. Last Update Accessed September 15, 2015.
1382 Montomoli E et al (2011) Current adjuvants and new perspectives in vaccine formulation. Expert Rev Vaccines 10: 1053-
1061
1383 Food and Drug Administration. Vaccine Product Approval Process.
http://www.fda.gov/BiologicsBloodVaccines/DevelopmentApprovalProcess/BiologicsLicenseApplicationsBLAProcess/ucm
133096.htm. Last Update 24 August 2015. Accessed 14 September 2015.
1384 Gruber M. Regulatory Pathways Supporting Development and Approval of Vaccines Formulated with Novel Adjuvant:
Regulatory Considerations and Challenges.
http://www.fda.gov/downloads/EmergencyPreparedness/MedicalCountermeasures/UCM292045.pdf. Last Update 2012.
Accessed 14 September 2015.
1385 Food and Drug Administration. Guidance for Industry: Clinical Data Needed to Support the Licensure of Seasonal
Inactivated Influenza Vaccines.
http://www.fda.gov/BiologicsBloodVaccines/GuidanceComplianceRegulatoryInformation/Guidances/Vaccines/ucm074794.
htm. Last Update 31 May 2007. Accessed 15 September 2015.
1386 Novartis Vaccines and Diagnostics. FDA Advisory Committee Briefing Document: Fluad Seasonal Adjuvanted Trivalent
Influenza Vaccine (aTIV).
http://www.fda.gov/downloads/AdvisoryCommittees/CommitteesMeetingMaterials/BloodVaccinesandOtherBiologics/Vacc
inesandRelatedBiologicalProductsAdvisoryCommittee/UCM461917.pdf. Last Update 15 September 2015. Accessed 21
September 2015.
1389 Ibid.

licensing delays the ability of this technology to impact influenza vaccination systems in the US in the
near term (i.e. within the next few years). Additionally, unless WHO strain selection meetings are delayed
to match the shorter production timescales of alternative platforms, these seasonal recombinant vaccines
will be subject to the same limitations due to strain selection and antigenic drift as egg/cell-based vaccines
(though are able to adjust production mid-stream if necessary, unlike egg/cell-based systems).1392
The benefits and limitations of GoF and alt-GoF approaches that shorten vaccine production timelines by
reducing the time needed for bulk antigen production are summarized in Table 15.11. It should be noted
that several other steps of the vaccine production process are time-consuming, such as preparation of
potency reagents for standardization of vaccine antigen and clinical trials (for pandemic vaccines). As
bulk antigen production times shrink, these other steps may become rate-limiting, unless new methods for
quantification of recombinant antigen are developed and FDA-approved.1393
1392 Ibid.
1393 Ibid.

Vaccine Development Benefits – Potential Benefits of Innovations that May Shorten Production Timelines for Strain-Specific Vaccines
Scientific/technical innovation Benefit Limitations Barriers
• Gains for high-yield CVVs are limited

by the production capacities of egg
Shorten production and cell systems
GoF: Improve yields of CVVs timelines for egg- and cell- o Egg-based production systems are • Likely none – minor modifications to
used for production of egg- and based vaccines by already near maximum levels of existing CVVs are unlikely to require FDA
cell-based vaccines increasing rates of bulk productivity approval for use in vaccine production
antigen production
• Several stages of egg/cell-based
production are time-consuming and
may become rate-limiting
• Gains for high-yield CVVs are limited

by the production capacities of egg
Shorten production and cell systems
Alt-GoF #1: Develop new host timelines for egg- and cell- o Egg-based production systems are
cell lines that permit higher levels based vaccines by already near maximum levels of • New cell lines must be FDA-licensed prior
of virus replication. increasing rates of bulk productivity to their commercial use.
antigen production
• Several stages of egg/cell-based
production are time-consuming and
may become rate-limiting
• Only one adjuvanted seasonal • Adjuvanted vaccines are considered “new”

vaccine (approved for use in adults and must be FDA-licensed prior to
Enable production of the aged 65 and older) and one commercial release
Alt-GoF #2: Use of adjuvants for same number of vaccine adjuvanted pandemic vaccine are o Development and licensing of new
antigen sparing doses in a shorter time FDA-licensed vaccines is a lengthy and expensive
period • Several stages of egg/cell-based process (requires clinical trials)
production are time-consuming and o More than five years, even using
may become rate-limiting accelerated regulatory pathway

Vaccine Development Benefits – Potential Benefits of Innovations that May Shorten Production Timelines for Strain-Specific Vaccines
Scientific/technical innovation Benefit Limitations Barriers
• Only one alternative vaccine (Flublok, • Development and licensure of new
a recombinant vaccine) is FDA- influenza vaccines is a lengthy and
licensed expensive process
o Others are in late stages of • Alternative standardization assays that do
Alt-GoF #3: Develop new vaccine development
Shorten production not depend on FDA-generated potency
platforms with faster production
timelines for strain- • Preparation of potency reagents for reagents must be FDA-licensed
timelines, such as recombinant flu
specific vaccines standardization of vaccine antigen is
vaccines • Will be subject to limitations associated
time-consuming and may become with strain selection far in advance of flu
rate-limiting season unless WHO strain selection
o Alternative standardization assays meetings are delayed to match shorter
could be used production timescales

15.2.4.3.3 Benefits of Alternative Approaches with Potential to Improve the Availability of Pandemic
Influenza Vaccines through Different Mechanisms
Because each of the GoF and alt-GoF approaches described above involves initiation of manufacturing
following the start of the pandemic, none can address the gap in protection in the immediate aftermath of
emergence of a novel strain. Several alternative approaches aim to proactively protect the public against
influenza pandemics, namely development of universal vaccines and development and stockpiling of pre-
pandemic vaccines.
A universal or broad-spectrum flu vaccine would obviate the need for production of a strain-specific
vaccine in response to the emergence of a novel pandemic strain. Such a vaccine could be administered in
advance of a pandemic, generating pre-existing immunity in the population, or could be stockpiled and
immediately deployed following the start of a pandemic. However, development of a universal or broad-
spectrum vaccine represents a scientifically challenging prospect. Although multiple research efforts are
underway, influenza and vaccinology experts disagree about whether a universal flu vaccine is
achievable, and one expert felt that a 10 to 20 year time frame for development of a universal vaccine is
optimistic.1394,1395,1396
Development of pre-pandemic vaccines against circulating zoonotic influenza strains with pandemic
potential would also lead to faster vaccine availability during a pandemic caused by a closely related
strain. Developing pre-pandemic CVVs and carrying out clinical trials would shorten vaccine production
timelines, and stockpiling bulk antigen would allow for near-immediate deployment of vaccine following
emergence of a pandemic strain. In addition, manufacturers’ experience with production of the vaccine
would likely streamline subsequent large-scale production during the pandemic.1397 Although the pre-
pandemic vaccine strain is unlikely to exactly match the strain that emerges to cause a pandemic, use of
adjuvants and prime-boost regimens broaden the protection that can be achieved using a strain-specific
vaccine, such that pre-pandemic vaccines are highly likely to provide some level of protection against
infection with a similar strain.1398,1399,1400,1401,1402
The benefit of developing pre-pandemic vaccines is constrained by the fact that resources for the
development and stockpiling of pre-pandemic vaccines are limited.1403 The number of pre-pandemic
CVVs that can be produced is constrained by two factors: (1) the number of facilities that can produce
pre-pandemic CVVs using Good Manufacturing Processes (GMP) is limited, and (2) CVVs used for
vaccines 7: 10-11
1396 (2015v) Influenza Vaccines. Interviews with Influenza Researchers.
1398 Ibid.
1399 (2015s) Influenza Vaccines. Interviews with Public Health Professionals Involved in Preventing and Responding to

vaccine production must undergo extensive safety and characterization testing, which is resource-
intensive.1404 Further along in the pre-pandemic vaccine production pipeline, the expense associated with
production of clinical lot material, clinical trials, and stockpiling of bulk antigen practically limits the
number of pre-pandemic vaccines that can be taken to each stage of the pipeline and the quantity of bulk
antigen that can be stockpiled.1405
The strengths and limitations of different strategies for improving the availability of pandemic influenza
vaccines are summarized in Table 15.12. Taken together, this analysis demonstrates that universal
vaccines are not a viable option for protection against pandemic influenza in the near future, but that
development of pre-pandemic vaccines represents a promising strategy due to their ability to provide
broad-spectrum protection when adjuvanted. However, because resources limit the scope of the USG’s
investment in pre-pandemic vaccines, these vaccines will serve to bridge the gap between the emergence
of a novel strain and widespread availability of vaccines and must be complemented by innovations to
shorten vaccine production timelines. Though one of several approaches that can achieve this benefit,
GoF research to improve CVV yields represents the only strategy for achieving near-term benefits
because it capitalizes on existing infrastructure and faces fewer regulatory barriers to translation than
other approaches.

Vaccine Development Benefits – Innovations that can Improve the Availability of Pandemic Influenza Vaccines
Scientific/technical Benefit Limitations Barriers
innovation
Scientifically challenging – influenza experts disagree
Alt-GoF: Universal or Population will already have immunity Universal and broad-spectrum about whether development of a universal vaccine is
broad-spectrum against novel strains that emerge and/or influenza vaccines do not yet feasible
influenza vaccine can be immediately vaccinated exist • One influenza vaccine production expert estimates
that a 10 – 20 year time frame is optimistic
Shorten production timelines
GoF or alt-GoF: • Refine vaccine formulation through
Invest in pre-pandemic clinical trials Pre-pandemic vaccine strain unlikely to match the
vaccine development* Scope of pre-pandemic strain that emerges
• Immediate availability of stockpiled vaccine development limited
• CVVs, clinical antigen by availability of funds • Use of adjuvants and prime/boost regimens broaden
trials, stockpiling of protection
• Manufacturing experiences
bulk antigen
facilitates subsequent large-scale
production
• Only one recombinant influenza vaccine is currently

Alt-GoF: Develop new vaccine Shorter vaccine production FDA-approved
platforms with faster production timelines would enable earlier
timelines release of vaccine • Other alternative vaccine platforms are in
development
Shorten production Would enable production of

timelines for strain- the same number of doses in a
Alt-GoF: Use of adjuvants for antigen Only one seasonal and one pandemic vaccines with
specific vaccines shorter period of time,
sparing adjuvants are licensed
enabling earlier vaccine
release
GoF or alt-GoF: Shorten production Shorter vaccine production
Gains for high-yield CVVs are limited by the
timelines for egg- and cell-based timelines would enable earlier
production capacities of egg and cell systems
vaccines* release of vaccine
*Both GoF and alt-GoF approaches can inform this benefit.

15.2.4.3.4 Benefits of Alternative Approaches with Potential to Improve the Efficacy of Seasonal
Influenza Vaccines through Different Mechanisms
Ultimately, GoF research and alternative approaches that shorten vaccine production timelines will
benefit public health during seasonal flu epidemics by enabling strain selection closer to the start of the
target flu season, which increases the likelihood that the vaccine strains will match circulating strains.
However, with the exception of universal flu vaccines, none of the strategies described above can
eliminate the need to choose vaccine strains in advance of flu season, thus vaccine mismatch remains a
possibility unless other innovations are pursued in tandem. Several other approaches have potential to
improve vaccine match through alternative mechanisms.
A universal or broad-spectrum vaccine would benefit public health responses to seasonal flu epidemics by
obviating the need for yearly production of strain-specific vaccines, but this strategy represents a
challenging, long-term approach.
As vaccine mismatch is sometimes due to incorrect prediction of which strains will predominate six to
nine months hence, improving strain selection capabilities represents another approach to increasing the
likelihood of vaccine match. Both GoF and alt-GoF approaches can improve strain selection capabilities,
described in detail in section 15.5.5.1 and briefly summarized here. First, both GoF and alt-GoF
approaches have potential to improve the quality and quantity of the antigenic characterization data upon
which strain selection decisions are based, thereby strengthening the robustness of the decision-making
process. Second, GoF approaches are critical for advancing the development of methods for predicting
antigenic drift, including experimental methods and computational methods. These methods would enable
production of vaccines based on future, antigenically drifted strains, which will match circulating viruses
at the time of vaccine deployment. Collectively, the benefits that can be achieved through both GoF and
alt-GoF approaches depend on scientific advancements as well as expansion of sequencing capabilities at
diagnostic labs that originally collect and characterize clinical samples. Both barriers will be challenging
to overcome, though small improvements to the state of the science and to surveillance infrastructure will
yield benefits. Thus the timescale for realization of these benefits is uncertain.
The strengths and limitations of different strategies for improving the efficacy of seasonal influenza
vaccines are summarized in Table 15.13. Universal vaccines represent the only strategy with potential to
fully “solve” the vaccine mismatch problem but are in early stages of development and represent a long-
term solution at best. All other approaches hold promise for improving the likelihood of vaccine match in
the near future. These approaches are complementary; that is, each approach addresses different
underlying gaps in current scientific and technical capabilities that contribute to vaccine mismatch. Thus,
these approaches complement each other as part of comprehensive strategy for improving the quality of
seasonal influenza vaccines. This includes GoF research that improves the yields of CVVs, thus
shortening vaccine production timelines, as well as GoF research that leads to the development of
genetically stable CVVs that retain the antigenicity of the parental strains.

Table 15.13. Innovations that can address seasonal influenza vaccine gaps associated with long vaccine production times
Gap Scientific/technical innovation Limitations Barriers
Scientifically challenging – influenza experts
disagree about whether development of a
Alt-GoF: Universal or broad- Universal and broad-spectrum influenza universal vaccine is feasible
Vaccine strains are often spectrum influenza vaccine vaccines do not yet exist • One influenza vaccine production expert
imperfectly matched to estimates that a 10 – 20 year time frame is
circulating strains optimistic
GoF: Development of genetically Limited to reducing the likelihood of
FDA approval may be required for
stable CVVs that are antigenically vaccine mismatch for those vaccine strains
commercial use of new CVVs
similar to parental strains that drift as a result of production
GoF or alt-GoF:
Improve strain selection
capabilities:*
Incorrect strain selection: Depends on advancements in science and
an unexpected strain • Improve the quality and quantity Limited to reducing the likelihood of improvements to influenza surveillance
of antigenic characterization vaccine mismatch due to incorrect strain
rises to prominence in networks, the timescales of which are
data considered during strain selection
nature during the vaccine uncertain
production process selection decision
• Predict antigenic drift, enabling
production of vaccines using
drifted strains
Lengthy production • GoF or alt-GoF: Improve CVV

times for egg- and cell- yields* Cannot eliminate the need to choose
based vaccines • Alt-GoF: Use of adjuvants for vaccine strains in advance of flu season
Described above (Table 15.8)
necessitate strain dose-sparing • The possibility of vaccine mismatch due
selection six months in • Alt-GoF: Develop new, faster to incorrect strain selection remains
advance of flu season vaccine platforms
*Both GoF and alt-GoF approaches can inform this benefit.

15.2.4.3.5 Summary – Benefits of GoF Approaches relative to alt-GoF approaches for future influenza
vaccine production
Production of pandemic influenza vaccines
Both GoF approaches to improve CVV yields and alternative approaches have potential to reduce the
time lag between the emergence of a novel pandemic strain in human populations and the widespread
availability of a vaccine, thus reducing human morbidity and mortality during an influenza pandemic.
GoF approaches to generate higher-yield CVVs and to identify mutations that enhance the growth of
CVVs are uniquely capable of achieving this benefit in the immediate to near term because use of this
information capitalizes on existing infrastructure and faces no regulatory barriers to translation.
Developing new host cell lines that permit higher levels of virus replication represents an alternative
approach for increasing CVV yields, but cell lines used for vaccine production must undergo extensive
testing for FDA licensure and this approach is not less risky than working with viruses with enhanced
yields. Although adjuvanted vaccines and virus-free vaccines have shorter production timelines than
existing egg- and cell-based vaccines, the length and expense of licensure processes for new vaccines will
delay their widespread availability. Universal flu vaccines are in early stages of development, and
influenza and vaccinology experts disagree about the scientific feasibility of developing a universal
vaccine. The development of pre-pandemic vaccines represents a promising strategy due to their ability to
provide broad-spectrum protection when combined with adjuvants. However, because investments in pre-
pandemic vaccines are resource-limited and strains that emerge are unlikely to exactly match the vaccine
strain, this approach does not abrogate the need to produce vaccine during a pandemic but rather bridges
the gap between strain emergence and widespread vaccine availability, thus complementing other
strategies for shortening vaccine production timelines.
Production of seasonal influenza vaccines
Both GoF approaches and alt-GoF approaches have potential to improve the match between seasonal
influenza vaccines and strains that are circulating during flu season, thereby improving vaccine efficacy
and decreasing human morbidity and mortality associated with seasonal flu epidemics. Because poor
vaccine match arises from several different shortcomings in the current vaccine production system, this
benefit can be achieved through several different mechanisms. One strategy is shortening the time needed
to produce flu vaccines, which enables strain selection closer to the start of flu season. As described
above, GoF approaches that improve the yields of CVVs are uniquely capable of achieving this benefit in
the immediate to near term, though alternative approaches such as incorporating adjuvants into existing
vaccines and developing virus-free vaccine platforms have strong potential to achieve this benefit over
longer timescales.
A completely different strategy is to improve the production of strains that mutate to alter their
antigenicity upon growth in eggs or cells, such as H3N2 strains, resulting in the production of vaccines
that are poorly matched to the selected strains. GoF approaches are uniquely capable of generating high-
yield, genetically stable CVVs that do not acquire antigenicity-altering mutations during passage in eggs
or cells.
A third strategy for improving vaccine match is to improve strain selection capabilities, which would
reduce the likelihood of mismatch due to incorrect predictions of which strains will be dominant during
the forthcoming flu season. This benefit can be achieved by improving the quantity and quality of the
antigenic characterization data upon which strain selection decisions are based, as well as by developing
methods for prediction of antigenic drift, to enable developing of vaccines based on future, drifted strains
that match circulating strains at the time of vaccine deployment. The former benefit depends on

strengthening influenza surveillance networks, and in particular expanding viral sequencing capabilities,
and both benefits rely on scientific advancements. As a result, both the extent and timescales of these
benefits are uncertain. Importantly, as these approaches address different underlying gaps in existing
vaccine production systems, research in these areas has the potential to complement the benefits that can
be achieved through the application of GoF research to vaccine production.
15.3 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Enhances
Mammalian Adaptation and Transmissibility
15.3.1 Overview of Influenza GoF Landscape
enhance the transmissibility of influenza viruses in mammals, including approaches that enhance the
fitness or infectivity of viruses in mammalian cells or in animals, respectively, as well as approaches that
enhance the transmissibility of viruses in appropriate animal models. In this section, an overview of GoF
approaches in this phenotypic category is provided and the scientific outcomes and/or products of each
approach are described.
15.3.1.1 Serial passaging of viruses in mammalian cells or animals
Serial passaging of viruses in mammalian cells in laboratory animals selects for viruses with enhanced
growth in cells or enhanced infectivity to animals, respectively. This type of serial passaging experiment
involves “forced” passaging, meaning that the experimenter directly transfers infected material, in the
form of cell culture supernatant or homogenates of infected tissue, to the subsequent cell culture dish or
animal. Forced serial passaging is carried out for two purposes: (1) to identify mutations that arise during
adaptation of animal influenza viruses (i.e. avian and swine viruses) to mammals, which provides a
foundation for follow-up studies investigating the evolutionary mechanisms driving adaptation to
mammalian hosts and the mechanistic basis of mammalian adaptation, and (2) to develop an mouse model
for the study of a particular virus.
15.3.1.2 Serial passaging of viruses in mammalian cells or animals with selection for transmission
Serial passaging of viruses in animals with selection for transmission leads to the generation of viruses
with enhanced transmissibility in mammals. This type of serial passaging experiment can involve
selection for contact transmission, during which the primary (directly inoculated) and secondary hosts are
co-housed, or for airborne transmission, during which the primary and secondary hosts are separately
housed in special isolator cages that prevent direct contact between animals but allow for air exchange
between cages. These studies seek to identify mutations that are sufficient to enhance transmissibility,
which provides a foundation for follow-up studies that investigate the mechanistic basis of
15.3.1.3 Forward genetic screen to identify genetic traits that enhance the fitness/transmissibility of
viruses in mammals
fitness/transmissibility or comprehensive reassortment of parental gene segments from two viruses,
followed by characterization of the fitness or transmissibility of mutants in appropriate mammalian model
systems to select for mutant viruses with enhanced fitness/transmissibility. Sequencing emergent viruses
enables the identification of mutations or gene segments that enhance the fitness/transmissibility of

viruses, which provides a foundation for follow-up studies that investigate the mechanistic basis of
15.3.1.4 Targeted genetic modification of viruses to introduce traits that are expected to enhance
fitness/transmissibility in mammals

introduce genetic traits that are expected to enhance the fitness/transmissibility of viruses followed by
characterization of the fitness or transmissibility of mutants in appropriate mammalian model systems
may lead to the generation of viruses with enhanced fitness/transmissibility in mammals. This approach is
phenotypic trait, such as a preference for binding to α2,6 sialic acid receptors, contributes to the complex
phenotypes of mammalian adaptation or transmissibility, and (2) to confirm that a particular mutation or
gene segment is necessary and sufficient to enhance the fitness/transmissibility of viruses in appropriate
model systems. Notably, genetic traits that are associated with mammalian adaptation/transmissibility
may be discovered through GoF approaches or alt-GoF approaches. As above, this information provides a
foundation for follow-up studies investigating the mechanistic basis of mammalian adaptation and
transmissibility.
15.3.2 Overview of the potential benefits of GoF approaches that enhance the
fitness/transmissibility of influenza viruses
GoF approaches have potential to benefit several aspects of scientific knowledge about the ability of
animal influenza viruses to adapt to efficiently infect and transmit between humans. GoF approaches can
provide insight into: (1) whether animal influenza viruses can acquire the capacity for airborne
transmissibility between mammals, (2) the evolutionary mechanisms driving adaptation of animal
influenza viruses to efficiently infect and transmit between mammals, and (3) the mechanistic basis of
mammalian adaptation and transmissibility of animal influenza viruses.
transmissibility between mammals have the potential to inform the interpretation of wildlife, agricultural
animal, and public health surveillance information. Specifically, determining the presence (or absence) of
particular mutations or of amino acid substitutions at particular sites is one aspect of evaluating the risk
posed by circulating animal influenza viruses. Risk assessments based on evaluation of genetic
surveillance data, as well as other types of data, then inform decision-making related to public health
preparedness for novel influenza outbreaks, as discussed below.
15.3.2.3 Vaccines
GoF approaches have the potential to benefit the development of pre-pandemic vaccines. Specifically,
pandemic risk assessments, which can be informed by GoF research (see section 16.3.2.2), may trigger
the development of candidate vaccine viruses based on high-risk viruses, as well as subsequent stages of
the pre-pandemic vaccine production pipeline (e.g. manufacturing of clinical lot material, conducting
human clinical trials, and stockpiling vaccine).

A lack of knowledge about whether existing therapeutics will be effective against future pandemic strains
hampers preparedness planning. GoF-generated viruses that are transmissible between ferrets may mimic
pandemic variants of that HA subtype better than wild type viruses. Thus, testing whether existing
therapeutics are capable of mitigating disease caused by GoF strains could inform pandemic preparedness
planning. Researchers have also suggested that these experiments could stimulate the development of new
therapeutics, in the event that existing therapeutics are found to be ineffective against GoF strains.
However, the relevance and utility of this information is severely constrained by several sources of
uncertainty, including a lack of knowledge about whether ferret-transmissible viruses are more
transmissible in humans, whether laboratory-generated transmissible viruses behave similarly to those
that could arise in nature, and other factors. Given this uncertainty, dedication of resources to developing
therapeutics targeting hypothetical future pandemic viruses is unlikely. Thus, this putative benefit to the
development of therapeutics is not considered in this report.
transmissibility between mammals contribute to assessments of the pandemic risk posed by circulating
animal influenza viruses, which are based on genetic surveillance data and several other types of data
(e.g. epidemiologic data, phenotypic data, etc.). These assessments inform policy decisions related to
public health preparedness for novel influenza outbreaks, including whether to develop and publicize
messaging about risk factors for contracting animal influenza infections and practices for mitigating risks,
whether to enhance surveillance of animals, and whether to develop pre-pandemic vaccines.
Pandemic risk assessments inform prioritization of resources for pandemic preparedness. Specifically,
evaluating the relative risk posed by different influenza viruses helps decision-makers allocate limited
funds to pandemic preparedness efforts, such as the development of pre-pandemic vaccines targeting
high-risk viruses. This prioritization may improve the efficiency of government spending on influenza
pandemic preparedness. Economic benefits were not explicitly evaluated in this report.
In this section, the ability of GoF approaches to address three key outstanding questions related to
influenza virus adaptation and transmission in humans is evaluated:
• Can animal influenza viruses become transmissible between humans?
1406
New diagnostics for novel influenza viruses are typically real-time PCR assays which include two or three diagnostic

• How do animal influenza viruses adapt to and become transmissible between humans? What
selective pressures drive adaptation and the evolution of efficient transmissibility, and what is the
order of acquisition of new genetic/phenotypic traits that are needed for
adaptation/transmissibility?
• What is the mechanistic basis of adaptation and transmission in humans? What viral factors are
involved, and what phenotypic changes must occur in order for an animal influenza virus to adapt
to efficiently infect, cause disease, and transmit in mammals?
Viral fitness and transmissibility in any model system are complex phenotypes that arise through the
cumulative effects of multiple underlying phenotypes, such as specificity for a particular type of cell
surface receptor and the ability to replicate within a particular temperature range. Generally, the
biological process of acquiring efficient transmissibility in a new host species can be viewed as the result
of two interdependent steps. First, the virus must be able to infect a new host, which depends on
underlying traits that contribute to mammalian adaptation, and second, the virus must be able to get out of
the primary host and infect a secondary host. Because the property of transmissibility depends on
phenotypes underlying both adaptation and transmission and because similar experimental approaches are
used to study both complex phenotypes, GoF experiments that enhance adaptation and transmissibility are
discussed together in this section.
The evolutionary mechanisms driving adaptation of viruses to new hosts and the acquisition of efficient
transmissibility, as well as the underlying genetic and phenotypic traits that enable efficient infection and
transmission in human populations, are poorly understood. Several phenotypes have been shown to be
associated with mammalian adaptation and transmissibility, including a preference for HA binding to cell
surface receptors decorated with α2,6 sialic acid moieties (versus “avian-like” α2,3 sialylated receptors),
the ability of the viral polymerase complex to function at lower temperatures, and an increase in HA
stability. However, considerable gaps in knowledge remain about the molecular basis of each phenotype
and the role of each phenotype in adaptation/transmissibility, and as-yet-undiscovered viral factors and
phenotypic changes are likely to contribute to the acquisition of efficient transmissibility in mammals.
Furthermore, the potential for animal influenza strains to evolve efficient transmissibility in humans is not
understood.
15.3.3.1 Scientific Knowledge Gap 1: Can animal influenza viruses become transmissible between
humans?
15.3.3.1.1 Benefits and Limitations of GoF Approaches
Several GoF approaches can lead to the generation of transmissible viruses, including deliberate genetic
modification of viruses and serial passaging of viruses in animals with selection for transmission.
Collectively, these approaches definitively demonstrate that a virus can acquire the capacity to transmit
between laboratory animals in an experimental setting. Notably, this approach can be applied to strains
that have not yet caused infections in human populations as well as strains that have caused human
infections but do not yet efficiently transmit in humans. The key limitations of this approach are that
observations in animal models may not translate to humans and that the adaptive changes observed in the
laboratory may not be possible in nature.
15.3.3.1.2 Benefits and Limitations of alt-GoF Approaches
Characterizing the transmissibility of wild type isolates in representative animal models represents an
alternative approach for addressing whether animal influenza viruses display the capacity for transmission

between mammals. However, this approach is inherently reactive – that is, it can effectively answer
whether a virus is transmissible but cannot shed light on whether a virus has the potential to become
transmissible. As above, observations in animal models may not translate to humans.
GoF approaches are uniquely capable of proactively assessing the potential for any animal influenza
viruses to acquire enhanced fitness and transmissibility in mammals. Notably, the relevance of this
information for human populations depends on the suitability of animal models as well as whether
laboratory-acquired mutations can arise in nature, both of which are unknown (Table 15.14).

Table 15.14. Summary of the Benefits of GoF Approaches that Enhance Transmission in Mammals
Scientific Knowledge Benefits—Can a virus acquire efficient transmissibility in appropriate animal models?
a
GoF #1 [1] : Targeted genetic modification
to introduce genetic changes expected to
contribute to transmissibility • Determine whether virus can acquire the
capacity for transmission in appropriate • Narrow breadth – Results may not generalize to other virus
GoF #2 [2]: Forward genetic screen to strains
animal models
introduce genetic changes that may
contribute to transmissibility o Proactive – can be performed using • Artificiality – Adaptive changes observed in the laboratory
viruses that do not yet transmit between may not be likely or possible in nature
GoF #3 [3]: Serial passaging with selection humans in nature
for transmission
• Narrow breadth – Results may not generalize to other virus

strains
• Reactive – Limited to analysis of viral isolates that already
Alt-GoF #1 [3]: Characterization of wild • Determine whether virus is transmitted in exist in nature
type virusesb appropriate animal models o Results from single round of selection may not reflect
virus capacity for evolution of transmissibility for
strains that have not yet caused infections in human
populations or strains that have caused human
infections but do not readily transmit in humans
a
GoF and alt-GoF approaches are listed in numerical order. Numbers in brackets reference the order in the landscape tables (Supplementary Information).
b
Note that to date, animal-origin viruses that efficiently transmit between humans have not been discovered in nature. If available, characterization and
phenotypic analysis of wild type isolates represents a viable alternative approach.

15.3.3.2 Scientific Knowledge Gap 2: How do animal influenza viruses adapt to and become
Serial passaging of animal influenza viruses in appropriate animal models to select for mammalian
adaptation and transmission, a GoF approach, provides insight into the mechanisms underlying adaptation
to mammals and the evolution of transmissibility. This approach is flexible, in that the method of
passaging (i.e. by direct inoculation, direct contact transmission, or airborne contact transmission) and the
tissue source used for forced passaging can be adjusted to study different modes of transmission.
Sequencing of isolates at multiple stages of passaging enables determination of the order and rate of
acquisition of adaptive traits, and follow-up studies elucidate how those genetic and phenotypic changes
influence other viral phenotypes. Comparing the sequences and phenotypes of viral isolates from different
tissues, different time points during the course of infection, and between the primary (directly inoculated)
and the secondary hosts can provide additional insight into the tissue-dependence of adaptation, the rate
of intra- and inter-host adaptation, and the selection pressures and viral population dynamics during
transmission, respectively. Notably, the adaptive changes that occur in the lab environment under forced
selection may not be relevant or possible during natural evolution, may not mimic adaptation and
transmission in humans, and may selectively represent the evolutionary course possible for the limited
number of viruses studied.
Serial passaging, as well as the alt-GoF methods described below, provides information about the genetic
traits that are associated with the acquisition of enhanced fitness and transmissibility in mammals.
However, to confirm which of these changes are necessary and sufficient to enhance fitness and
transmissibility, targeted mutagenesis must be used to re-introduce mutations into parental strains
followed by characterization of the infectivity/transmissibility of mutant strains. Targeted mutagenesis
also enables determination of how the order of acquisition of genetic changes influences other viral
phenotypes, such as replicative fitness, which has implications for the likelihood that these traits can arise
in nature.
Several alt-GoF approaches can also address how influenza viruses evolve to efficiently infect and
transmit in humans. First, the comparison of sequences from closely related human and animal isolates
enables the identification of the origin and evolutionary rate of genetic changes among circulating viruses,
which can provide information on selection pressures and diversity among viruses in different hosts. That
this approach examines the natural course of adaptation and underlying mechanisms of infection and
transmission of viruses in humans is a strength relative to GoF approaches and other alternatives that
depend on the suitability of animal models in an artificial environment as representative of human
disease. An additional strength of the comparative sequence analysis method is the ability to analyze
genetic features across broad data sets including many viral isolates
This approach suffers from several significant limitations. The use of comparative sequence analysis is
feasible only if human-adapted and transmissible viruses have arisen in nature, but to date, animal
influenza viruses have limited capacity to infect and transmit in humans. Analysis of the few animal-
origin spillover infections may however inform evolution of adaptive traits. The success of this approach
depends on the quality and availability of surveillance data. In particular, the noisiness of comparative
sequence analysis due to high genetic diversity among influenza viruses practically limits this approach to
the examination of genetic regions known to be important for adaptation and transmissibility. The

identification of precursor strains that are closely related to zoonotic or human-adapted viruses
strengthens the utility of this approach by reducing the genetic diversity between compared strains,
however precursor-spillover paired strains have not been identified in all cases. Moreover, available
sequences may not capture all of the critical adaptive steps and cannot identify traits that were lost or
negatively selected during adaptation (i.e. evolutionary pathways not taken). Thus this approach may
provide less depth of information about how positively and negatively selected genetic traits interact to
determine fitness in distinct host populations and during transmission.
Analysis of viruses that have emerged from avian or mammalian reservoirs to become transmissible in
other mammalian species represents another surveillance-based approach for studying the mechanisms
underlying adaption to mammals during interspecies transmission. The recent emergence of animal
transmissible influenza viruses in other mammals (e.g. an avian-origin H3N2 canine influenza virus that
emerged in dogs in the mid-2000s) enables the study of the full evolutionary pathway for cross-species
acquisition of efficient transmissibility. This approach is subject to the same limitations as comparative
sequence analysis of human and animal isolates, with the additional caveat that adaptation to other
mammals may occur through different pathways and mechanisms than in humans.
Phenotypic characterization of wild type viruses by evaluating infectivity and transmissibility in

appropriate model systems is another alt-GoF approach for studying the evolution and mechanisms of
adaptation/transmissibility. This approach allows for the generation of detailed information about intra-
and inter-host evolutionary dynamics and can uncover both negatively and positively selected mutations.
However, it is limited to observation of adaptive changes over a single round of transmission, effectively
limiting the time and selective pressure under which adaptation occurs. It should be noted that in some
cases, mutations associated with adaptation and transmissibility can be generated in vivo within a single
round of transmission. Any animal influenza viruses that are highly attenuated in representative animal
models or are incapable of establishing infection are not suitable for this approach. Furthermore, this
approach is limited by its narrow breadth and depends on the suitability of the animal models used for
characterization.
section, with respect to the ability of each approach to provide insight into the evolution of fitness and
transmissibility in mammals, are summarized in Table 15.15. Taken together, GoF approaches are
uniquely capable of providing in-depth information about the evolution of mammalian
fitness/transmissibility in any animal influenza virus strain. In addition, GoF approaches are uniquely
capable of demonstrating the order(s) of acquisition of genetic changes that are necessary and sufficient to
lead to enhanced fitness/transmissibility in mammals. However, the relevance of information derived
from GoF approaches is contingent upon how well animal models represent human disease and how well
the lab environment mimics natural evolution.
For those wild type strains that are naturally capable of productively infecting laboratory animals used for
transmission studies, simply characterizing the transmissibility of a strain in animals, an alt-GoF
approach, has the potential to generate similarly in-depth information. However, a single round of
transmission may be insufficient for relevant adaptive changes to accrue or may reveal only part of the
adaptive process, which further lessens the relative utility of this alt-GoF approach. Surveillance-based
approaches, including comparison of human and animal isolates and comparison of animal isolates from
different species, are uniquely capable of reporting on the real-world evolution of a variety of strains, thus
complementing two shortcomings of GoF approaches. Though results gleaned from comparative analysis
of human and animal isolates are directly translatable to humans, the fact that animal influenza virus
strains that efficiently transmit in humans have not been observed in nature precludes use of this approach

for the study of transmissibility in particular. While case studies of interspecies transmission events exist,
the translatability of that information to the evolution of human adaptive traits is uncertain.

Scientific Knowledge Benefits—How do animal-origin viruses adapt to and become transmissible in humans?
• Provide insight into evolutionary mechanisms driving

adaptation/transmissibility
• Determine order of acquisition of genetic changes that
GoF #1 [1]a: Targeted genetic modification to
introduce genetic changes expected to
are necessary and sufficient to enhance • Translatability – Results may not translate to adaptation in
adaptation/transmissibility humans
contribute to adaptation/transmissibility o Determine how rate and order of acquisition of
genetic changes affects other viral phenotypes
• Proactive - can be performed using viruses that do not
yet transmit between humans in nature
• Provides in-depth insight into evolutionary
mechanisms driving adaptation/transmissibility
o Captures all adaptive steps
o Identifies positively and negatively selected traits • Narrow breadth – Results may not generalize to other virus
o Evaluates adaptation over a long time period and strains
GoF #2 [3]: Serial passaging with or without under high selective pressures • Artificiality – Adaptive changes observed in the laboratory
selection for transmission • Determine how rate and order of acquisition of genetic may not be representative of evolution in nature
changes affects other viral phenotypes • Translatability – Results may not translate to adaptation in
• Proactive - can be performed using viruses that do not humans
yet transmit between humans in nature
• Uncovers previously unidentified genetic and
phenotypic traits mediating evolution

• Lack of correlateb – Animal-origin viruses have limited
capacity to infect and transmit in humans, limiting the
availability of suitable data
• Bias – High genetic diversity among influenza strains limits
this approach to the examination of genetic regions already
• Provide insight into evolutionary mechanisms driving known to be important
o Identify the origin and evolutionary rate of genetic
• Limited/poor foundational knowledge of the linkage between
changes among circulating viruses, underlying phenotypes and adaptation/transmissibility provide
Alt-GoF #1 [1]: Comparative sequence further uncertainty
analysis of human and animal isolates o Provides information on the natural evolutionary
process, • Limited by the quality and availability of existing surveillance
o Directly translates to human adaptation and data
disease, and o Consensus sequences may not capture extent of viral
o Analyzes broad data sets applicable to many strains diversity
• Static – Cannot identify lost or negatively selected traits, and
intermediate adaptive events may not be captured
• Associative – Information produced is correlative, not
causative
• Bias – High genetic diversity among influenza strains limits
this approach to the examination of genetic regions already
known to be important
underlying phenotypes and adaptation/transmissibility provide
further uncertainty
• Provide insight into evolutionary mechanisms driving • Limited by the quality and availability of existing surveillance
adaptation/transmissibility data
Alt-GoF #2 [2]: Comparative sequence o Identify the origin and evolutionary rate of genetic o Consensus sequences may not capture extent of viral
analysis of animal isolates from two species changes among circulating viruses, diversity
o Provides information on the natural evolutionary • Static – Cannot identify lost or negatively selected traits, and
process intermediate adaptive events may not be captured
o Analyzes broad data sets applicable to many strains
causative
• Translatability – Results may not translate to adaptation in
humans
o Whether animals under study are representative models for
human disease has not been established

• Narrow breadth – Results may not generalize to other virus
strains
• Provides insight into evolutionary mechanisms driving • Translatability – Results may not translate to adaptation in
adaptation/transmissibility humans
o Captures all adaptive steps • Associative – Information produced is correlative, not
o Identifies positively and negatively selected traits causative
Alt-GoF #3 [3]: Characterization of wild type • Determine how rate and order of acquisition affects • Limited to use of viruses that can productively infect
viruses other viral phenotypes representative animal models
• Proactive - can be performed using viruses that do not • Single (vs. multiple) round of infection/transmission limits the
yet transmit between humans efficiently in nature time for evolution and the amount of applied selection
• Uncovers previously unidentified genetic and pressure
phenotypic traits mediating evolution o May be insufficient time fro relevant evolutionary changes
to accrue
o May capture only part of the adaptive process
aGoF and alt-GoF approaches are listed in numerical order. Numbers in brackets reference the order in the landscape tables (Supplementary Information).
b Notethat to date, animal-origin viruses that efficiently transmit between humans have not been discovered in nature. If available, comparative sequence and phenotypic
analysis represents a viable approach for identification of genetic markers associated with human adaptation/transmissibility.

15.3.3.3 Scientific Knowledge Gap 3: What are the genetic and phenotypic traits that result in adaption
Several GoF approaches can be used to discover the genetic and phenotypic markers underlying
mammalian adaptation and transmission of animal influenza viruses include:
to adaptation and transmission in mammals by either site-directed mutagenesis or targeted
reassortment (often between animal and human seasonal strains),

by selection for mammalian infectivity, transmissibility, or underlying phenotypes, and
• Serial passaging in appropriate animal models or mammalian cells to select for mammalian
adaptive or transmissible traits.
enhanced fitness in cell culture model systems or infectivity and transmissibility in animal models, which
provides a foundation for follow-up biochemical, cell biological, and structural studies that elucidate
associated phenotypic changes. Serial passaging has the potential to uncover novel genetic and phenotypic
markers that contribute to adaptation/ transmissibility. In contrast, because forward genetic screens
involving random mutagenesis typically focus on regions that are suspected or known to play a role in
phenotypes underlying adaptation/transmissibility, this approach can discover novel genetic markers for
adaptation/transmissibility only. The targeted genetic modification approach is limited to the investigation
of genetic traits and underlying phenotypes that are suspected to contribute to adaptation/transmissibility
(e.g. determining whether altering sialic acid receptor binding specificity contributes to transmissibility).
necessary and sufficient to enhance infectivity or transmissibility in mammals. The use of in vitro model
systems is limited to the investigation of phenotypes underlying adaptation and transmissibility, such as
replicative fitness and sialic acid receptor specificity. Moreover, the results derived from these studies
may not be recapitulated in the complex environmental pressures encountered in a host. The relevance of
both in vitro and in vivo approaches depends on whether mechanisms underlying adaptation to cell culture
and animal models are representative of those in humans, and results gleaned from the study of one or a
few strains may not be recapitulated in different genetic contexts.
Several alt-GoF approaches can be used to uncover genetic and phenotypic traits underlying adaptation
and transmission in mammals. First, comparing the sequences of human and animal isolates enables the
identification of genetic changes that are associated with human adaptation and transmissibility. Unlike
the GoF approaches described above, this approach has the potential to directly identify human-adaptive
traits and may be more likely to uncover conserved traits through analysis of a large number of strains.
However, the fact that no animal influenza viruses that efficiently transmit in humans have been observed
in nature precludes the use of this approach to identify mechanisms underlying transmissibility. For the
discovery of mammalian adaptive traits, the success of this approach depends on the quality and
availability of surveillance data. In particular, the fact that nearly all published sequences represent
consensus sequences means that the presence of rare adaptive traits that arise in human cases may not be
captured in the data. Finally, the extensive genetic diversity within circulating virus populations and

among viruses isolated from humans makes discerning distinct genetic traits that are likely to contribute
to fitness and transmissibility in humans relative to animals difficult. Namely, the “noise” associated with
sequences comparisons obscures the discovery of relevant features that distinguish human versus animal
isolates, which practically limits this approach to the investigation of traits or regions previously known
to be important for adaptation.
Comparing the sequences of evolutionarily related isolates from different animal species represents
another surveillance-based approach for identifying genetic traits that are associated with mammalian
adaptation and transmissibility. Importantly, because avian-origin flu viruses that are airborne or contact
transmissible exist in circulation in several mammals including seals, horses, and dogs, this approach is
currently feasible for the study of transmissibility. In addition to the limitations above, mechanistic
insight gleaned through this approach may not translate to the adaptation of animal influenza viruses to
humans.
Phenotypic characterization of wild type viruses in appropriate animal models is another alt-GoF
approach that complements the use of surveillance data to study mechanisms underlying mammalian
adaptation and transmissibility. Specifically, comparing the sequences of wild type viruses with varied
levels of fitness and transmissibility enables the identification of genetic traits associated with
fitness/transmissibility. This approach is limited to the study of viruses that can productively infect and
transmit between animal models for adaptation/transmission. Notably, very few natural animal-origin
viruses are capable of transmission in ferrets and many are not able to efficiently cause disease in
representative animal models. Similarly to GoF approaches, genetic and phenotypic traits uncovered
through this approach may not translate to human-adapted viruses and may only be applicable to the
limited number of strains analyzed.
modification to identify genetic changes that attenuate fitness and transmission in mammals, can provide
information about genetic and phenotypic traits that contribute to transmissibility. Targeted LoF can also
be used to confirm necessary genetic or phenotypic traits by determining that mutations attenuate fitness
or transmission, but cannot identify traits that lead to enhanced transmission. This approach suffers from
several significant limitations. First, LoF studies can be performed only using transmissible seasonal or
pandemic viruses, and insights may not translate to animal influenza viruses. Second, because of the high
mutation rate of influenza viruses, LoF mutations that attenuate transmissibility may revert during the
single round of passage that is needed to characterize the transmissibility of the mutants (which represents
a selection step). Third, because many mutations attenuate transmission for trivial reasons, for example
mutations that compromise viability, discovering traits that directly contribute to transmissibility may be
difficult using a LoF approach. Finally, although in principle LoF screens can be performed after random
mutagenesis to discover new genetic elements important for transmission, the resource intensive nature of
transmission studies in ferrets practically limits these studies to the investigation of a few, known targets.
Several in vitro virus-free methods can be used to investigate phenotypes underlying adaptation and
transmissibility. Comparative sequence analysis of viral proteins with different phenotypic properties can
then enable the identification of mutations that are associated with relevant phenotypic changes, while
forward genetic screens can be used to identify novel genetic traits that contribute to underlying
phenotypes. Additional characterization involves the use of biochemical assays (e.g. characterizing the
acid stability of the HA protein) and crystallographic resolution of the structures of virus-host protein
complexes can provide insight into the functional and biophysical basis of underlying phenotypes. The
use of targeted modification of viral gene segments in isolation can also effectively confirm the necessary
and sufficient genetic traits that alter an underlying phenotype. Though the simplicity and relatively high-
throughput nature of these methods renders them appealing as a screening approach for the discovery and
confirmation of novel genetic traits that contribute to adaptation/transmissibility, these approaches are

inherently limited to the characterization of phenotypes (and genetic traits in the case of targeted
modification) previously identified in other experiments. An additional drawback is that results gleaned
from studying the behavior of a viral protein or phenotype in isolation may not be recapitulated in the
context of the full virus or in vivo. Moreover, although fairly rapid phenotypic assays have been
developed for the study of phenotypic traits known to be associated with adaptation/transmissibility,
assays to study phenotypic traits may be unreliable or unavailable for future phenotypes of interest.
Structure-based modeling approaches, an in silico method, may also be used to predict the effects of
mutations on phenotypes underlying adaptation/transmissibility. This approach is critically limited by the
capabilities and accuracy of existing models, and as such any conclusions may not be consistent in the
context of the full virus.
Finally, several alt-GoF approaches focus on identifying host factors and host-virus interactions that are
associated with mammalian adaptation, which may provide indirect insight into viral mechanisms
underlying cross-species adaptation. Specifically, in vitro proteomic (e.g. mass spectrometry) and
genomic screens (e.g. RNAi screen) utilizing both virus-free and cell culture-based infection systems are
used to identify host factors that interact with virus proteins of interest or that are critical for underlying
phenotypes such as viral replication. These approaches complement the identification of viral
proteins/phenotypes underlying adaptation to new hosts. However, the breadth of proteomic approaches is
limited in that screens typically focus on a single viral protein, and both genomic and proteomic screens
can identify host proteins that may not be functionally relevant or may play minor roles in the viral life
cycle.
Another type of alternative approach involves the use of attenuated viruses for GoF methods, as a risk
mitigation strategy. Three types of attenuated viruses are used for such studies: reassortants with surface
protein gene segments from seasonal influenza viruses, to which the general population has pre-existing
immunity, reassortants with lab-adapted viruses (e.g. PR8), and strains which have virulence factors
altered or deleted (e.g. deletion of the multi-basic cleavage site in HPAI HA sequences). Results gleaned
through use of attenuated viruses may be of limited informational value because complex, multi-genic
traits depend on genetic context (a phenomenon called epistasis), and results may not be recapitulated in
the context of the wild type virus. Differences in disease pathogenesis, which critically influences the
biological processes of adaptation and transmission, further compromise the relevance of results gained
through the use of attenuated strains. In addition, several factors limit the range of information that can be
generated using attenuated strains. First, seasonal reassortant strains can be used to study the role of genes
that encode internal factors (e.g. polymerase and nucleoprotein, etc.) only, while lab-adapted reassortants
are limited to the study of proteins donated by the wild type strain. Second, lab-adapted reassortants
cannot cause disease or transmit in ferrets and thus cannot be used to study airborne transmissibility in
this model system. Other types of attenuated strains, such as strains in which the multi-basic cleavage site
has been deleted, may not be suitable for in vivo studies.
Tables 15.16 and 15.17 summarize the benefits and limitations of GoF and alt-GoF approaches that
provide insight into the mechanisms underlying the fitness and transmissibility of influenza viruses in
mammals. Taken together, GoF approaches are uniquely capable of identifying novel genetic and
phenotypic traits underlying mammalian adaptation and transmissibility in any animal influenza virus
strain of interest. Furthermore, targeted genetic modification of viruses to introduce genetic traits
associated with mammalian adaptation/transmissibility is uniquely capable of demonstrating that
particular genetic markers are necessary and sufficient for mammalian adaptation and transmissibility
across multiple virus contexts. Given the importance of genetic context for influenza biology, this
approach critically strengthens the certainty of scientific knowledge about mechanisms underlying

mammalian adaptation and transmissibility. However, results gleaned from cell culture and animal model
studies may not translate to human disease. Notably, attenuated strains cannot be used to study
mechanisms underlying airborne transmission because these strains do not efficiently infect ferrets.
Additionally, attenuation alters disease pathogenesis and compromises the utility of the information
gleaned through studies using other model systems.
Characterizing wild type viruses, an alt-GoF approach, also has the potential to uncover previously
unknown traits. However, the fact that this approach cannot be used to study animal influenza viruses that
do not productively infect laboratory animals and that relevant changes may not arise during a single
round of transmission renders it less useful than GoF approaches. LoF approaches have limited utility for
broad and unbiased identification of necessary genetic and phenotypic traits due to their inefficiency and
the fact that mechanisms underlying transmissibility of seasonal/pandemic viruses may not translate to
animal influenza viruses. The simplicity and relative high-throughput nature of in vitro, virus-free
systems renders them appealing for the discovery of novel genetic traits that alter known phenotypes
underlying mammalian adaptation/transmissibility, but properties observed may not be recapitulated
during the complete viral life cycle.
Unlike GoF methods, the use of human and animal surveillance data for the discovery of genetic markers
associated with adaptation/transmission directly translates to human disease and has strength in numbers
as it analyzes genetic traits across large data sets. Critically, this approach cannot be used for studying
transmissibility because animal or zoonotic viruses that efficiently transmit in humans have not been
observed in nature. Analysis of sequences spanning avian to mammalian adaptation events enables the
identification of “real-world” markers associated with mammalian adaptation/transmissibility but may not
translate to human-adapted viruses. For both surveillance-based approaches, shortcomings in the quality
and availability of surveillance data compromise the feasibility of this approach and the relevance of any
findings.
Finally, host-focused approaches, such as proteomic and genomic screens, cannot supplant the
identification of viral adaptation/transmissibility traits but rather complement GoF approaches by
identifying host factors that contribute to those processes.

Scientific Knowledge Benefits— What are the genetic and phenotypic traits of mammalian adaptation and transmissibility of influenza viruses?
• Identifies genetic and phenotypic traits that are

necessary and sufficient for adaptation to • Narrow breadth – Results may not generalize to other virus strains
mammals or enhanced transmissibility (i.e.
provides causative data) • Translatability – Results from representative animal models may not translate to
GoF #1a [1]a: Targeted mechanisms underlying transmissibility in humans
genetic modification to • Gain insight into phenotypes underlying
introduce genetic changes adaptation/transmissibility • Bias – Limited to investigation of previously identified phenotypic or genetic traits
expected to contribute to • Proactive - can be performed using viruses that • Limited/poor foundational knowledge of the linkage between underlying phenotypes
transmissibility (in vivo) and adaptation/transmissibility provide further uncertainty
do not yet transmit between humans in nature
• Enables testing of markers in different strain • Lack of publication of negative results – Compromises evaluation of whether the
contexts to assess generalizability of previous function of particular markers is broadly conserved
findings
• Identifies genetic and phenotypic traits that are • Limited to the investigation of viral fitness, which is one component of mammalian
necessary and sufficient for viral fitness (i.e. adaptation and transmissibility
provides causative data)
• Narrow breadth – Results may not generalize to other virus strains
GoF #1b [1]: Targeted • Gain insight into phenotypes underlying fitness
genetic modification to • Translatability – Results from cell culture models may not translate to humans
introduce genetic changes • Proactive - can be performed using viruses that
• Bias – Limited to investigation of previously identified phenotypic or genetic traits
expected to contribute to do not yet transmit between humans efficiently in
transmissibility (in vitro) nature • Limited/poor foundational knowledge of the linkage between underlying phenotypes
and adaptation/transmissibility provide further uncertainty
• Enables testing of markers in different strain
contexts to assess generalizability of previous • Lack of publication of negative results – Compromises evaluation of whether the
findings function of particular markers is broadly conserved
• Identifies novel genetic traits that are sufficient • Narrow breadth – Results may not generalize to other virus strains
for mammalian adaptation/enhanced
GoF #2a [2]: Forward
transmissibility
• Translatability – Results from representative animal models may not translate to
genetic screen to introduce mechanisms underlying transmissibility in humans
genetic changes that may • Gain insight into phenotypes underlying
• Bias – Limited to investigation of previously identified phenotypic traits
contribute to adaptation/transmissibility
transmissibility, followed
• Proactive - can be performed using viruses that
• Limited/poor foundational knowledge of the linkage between underlying phenotypes
by testing in vivo and adaptation/transmissibility provide further uncertainty
do not yet transmit between humans efficiently in
nature • Associative – Information produced is correlative, not causative


• Identifies novel genetic traits that are sufficient to
enhance viral fitness • Limited to the investigation of viral fitness, which is one component of mammalian
GoF #2b [2]: Forward adaptation and transmissibility
genetic screen to introduce • Gain insight into phenotypes underlying fitness
genetic changes that may • Translatability – Results from cell culture models may not translate to humans
contribute to phenotypes
• In vitro methods can be used to screen a larger
underlying transmissibility,
number of mutants than in vivo methods • Bias – Limited to investigation of previously identified phenotypic traits
followed by testing in vitro • Proactive - can be performed using viruses that • Limited/poor foundational knowledge of the linkage between underlying phenotypes
do not yet transmit between humans efficiently in and adaptation/transmissibility provide further uncertainty
nature
• Associative – Information produced is correlative, not causative
• Identifies novel genetic and phenotypic traits that

are sufficient for mammalian
GoF #3a [3]: Serial
adaptation/enhanced transmissibility • Narrow breadth – Results may not generalize to other virus strains
passaging with selection for • Gain insight into phenotypes underlying • Translatability – Results from representative animal models may not translate to
transmission, use of animal adaptation/transmissibility mechanisms underlying transmissibility in humans
models (in vivo)
• Proactive - can be performed using viruses that • Associative – Information produced is correlative, not causative
nature
• Identifies novel genetic and phenotypic traits that • Narrow breadth – Results may not generalize to other virus strains
GoF #3b [3]: Serial are sufficient to enhance viral fitness
• Limited to the investigation of viral fitness, which is one component of mammalian
passaging with selection for • Gain insight into phenotypes underlying fitness adaptation and transmissibility
transmission, use of cell
culture models (in vitro) • Proactive - can be performed using viruses that • Translatability – Results from cell culture models may not translate to humans
nature • Associative – Information produced is correlative, not causative

• Identifies genetic traits that are associated with

human adaptation/transmissibility
o Comparison of genetically similar viruses, such
as precursor strains/spillover pairs, can result in • Lack of correlateb – Animal-origin viruses have limited capacity to infect and transmit
the identification of previously unknown in humans, limiting available data
genetic traits that are associated with human • Bias – High genetic diversity among influenza strains limits this approach to the
adaptation examination of genetic regions already known to be important
o Depending on the size of analysis and strength
Alt-GoF #1 [1]: of association some traits can be considered
Comparative sequence and adaptation/transmissibility provide further uncertainty
“causally” linked
analysis of human and • Limited by the quality and availability of existing surveillance data
o Directly translates to human adaptation and
animal isolates
disease o Consensus sequences may not capture low frequency mammalian- adaptive
o Analyzes broad data sets applicable to many mutations
strains o High genetic diversity impairs identification of precursor strains
• Gain insight into the prevalence and distribution o Limited reporting of negative surveillance data
of genetic traits • Associative – Information produced is correlative, not causative
o Can infer functional generalizability of genetic
traits (i.e. whether they do or do not behave
similarly in different genetic contexts)

• Identifies genetic traits that are associated with • Lack of correlatea – Animal-origin viruses have limited capacity to infect and transmit
cross-species adaptation/transmissibility in humans, limiting available data
o Comparison of genetically similar viruses, such • Bias – High genetic diversity among influenza strains limits this approach to the
as precursor strains/spillover pairs, can result in examination of genetic regions already known to be important
the identification of previously unknown
genetic traits that are associated with • Limited/poor foundational knowledge of the linkage between underlying phenotypes
mammalian adaptation and adaptation/transmissibility provide further uncertainty
Alt-GoF #2 [2]: o Depending on the size of analysis and strength • Limited by the quality and availability of existing surveillance data
Comparative sequence of association some traits can be considered
analysis of animal isolates o Consensus sequences may not capture low frequency mammalian- adaptive
“causally” linked mutations
from two species
o Analyzes broad data sets applicable to many o High genetic diversity impairs identification of precursor strains
strains
o Limited reporting of negative surveillance data
• Gain insight into the prevalence and distribution
of genetic traits
o Can infer functional generalizability of genetic • Translatability – Results may not translate to adaptation in humans
traits (i.e. whether they do or do not behave o Whether animals under study are representative models for human disease has not
similarly in different genetic contexts) been established
• Lack of correlatea – Animal-origin viruses may have limited capacity to infect animal
• Identifies genetic and phenotypic traits that are models and have a highly limited capacity to transmit in animal models, limiting
associated with mammalian suitable isolates for this approach
adaptation/transmissibility • Bias – High genetic diversity among influenza strains limits this approach to the
Alt-GoF #3 [3]: o Comparison of genetically similar viruses, such examination of genetic regions already known to be important
Characterization of wild as precursor strains/spillover pairs, can result in
type viruses the identification of sufficient genetic and • Limited/poor foundational knowledge of the linkage between underlying phenotypes
phenotypic traits and adaptation/transmissibility provide further uncertainty
• Gain insight into phenotypes underlying • Translatability – Results may not translate to mechanisms underlying
adaptation/transmissibility fitness/transmissibility in humans

• Translatability – Results may not translate to mechanisms underlying

fitness/transmissibility in humans
Alt-GoF #4 [4]: LoF • Identifies previously unknown genetic and • Limited Range – Limited to investigation of transmissible seasonal or pandemic
forward genetic screen to phenotypic traits that are necessary for viruses;
introduce genetic changes mammalian adaptation/enhanced transmissibility
that may attenuate • Attenuated virus may recover transmissibility during characterization
transmissibility, followed • Gain insight into phenotypes underlying • Bias – Limited to investigation of previously identified phenotypic traits
by testing in vitro or in vivo adaptation/transmissibility
and adaptation/transmissibility provide further uncertainty
• Triviality – May uncover mutations that indirectly attenuate
adaptation/transmissibility, which provides limited mechanistic insight

necessary for mammalian • Limited Range – Limited to investigation of transmissible seasonal or pandemic
adaptation/transmissibility (i.e. provides causative viruses; results may not generalize to other influenza sub-types
Alt-GoF #5 [5]: Targeted data) • Attenuated virus may recover transmissibility during characterization
LoF to introduce genetic
changes expected to • Gain insight into phenotypes underlying • Bias – Limited to investigation of previously identified phenotypic or genetic traits
attenuate transmissibility adaptation/transmissibility
• Enables testing of markers in different strain and adaptation/transmissibility provide further uncertainty
contexts to assess generalizability of previous
findings • Triviality – May uncover mutations that indirectly attenuate
adaptation/transmissibility, which provides limited mechanistic insight
• Lack of publication of negative results – Compromises evaluation of whether the
function of particular markers is broadly conserved

• Identifies novel genetic traits that are sufficient to

alter phenotypes underlying
Alt-GoF #6 [6]: (In vitro,
virus-free) Forward genetic • Provides insight into mechanistic basis of
screen to introduce genetic underlying phenotypes
changes that may alter
phenotypes underlying
• In vitro methods can be used to screen a larger
number of mutants than in vivo methods
• Proactive - can be performed using virus gene
segments from viruses that do not yet transmit
between humans in nature • Simplicity of model system – Results based on the study of a viral protein/phenotype
in isolation may not be recapitulated in the context of the full virus
• Identifies genetic traits that are necessary and
sufficient to alter a phenotype underlying • Narrow breadth – Results may not generalize to other virus strains
Alt-GoF #7 [11]: (In vitro, • Provides insight into the mechanistic basis of
virus-free) Targeted genetic phenotypes underlying adaptation/transmissibility • Limited/poor foundational knowledge of the linkage between underlying phenotypes
modification to introduce and adaptation/transmissibility provide further uncertainty
genetic changes expected to • Proactive - can be performed using virus gene
alter phenotypes underlying segments from viruses that do not yet transmit • State of methodology – Relies upon phenotypic assays, which may be unreliable or
adaptation/transmissibility between humans in nature unavailable
• Enables testing of markers in different viral gene • Lack of publication of negative results – Compromises evaluation of whether the
segments to assess generalizability of previous function of particular markers is broadly conserved
findings
• Provides insight into biophysical mechanisms

underlying virus-host and virus-virus protein
Alt-GoF #8 [7]: (In vitro, interactions
virus-free) Structural
o Provides detailed mechanistic information
studies to analyze the
molecular basis of • Proactive - can be performed using select virus
adaptation/transmissibility gene segments from viruses that do not yet
transmit between humans in nature depending on
the state of methodology


Alt-GoF #9 [12]: (In silico, interactions
virus-free) Modeling to • Provides detailed mechanistic information • Predictive – Does not confirm or correlate phenotypic effects in a biological context
analyze the biophysical
effects of mutations • Proactive - can be performed on virus gene • Simplicity of model system – Results based on the study of a viral protein/phenotype
contributing to phenotypes segments from viruses that do not yet transmit in isolation may not be recapitulated in the context of the full virus
underlying between humans in nature • Model accuracy – Utility of the approach depends on the quality of existing models
• Enables prediction of phenotypic consequences of
markers in different viral gene segments to assess
generalizability of previous findings

Alt-GoF #10 [8]c:  Identifies host proteins that may play a role in • Narrow breadth – Results may not generalize to other virus strains
Proteomic screen to identify mammalian adaptation during infection
host proteins that physically
o Reveals previously unknown host factors • Bias – Limited to investigation of previously identified phenotypic traits
interact with viral proteins
during infection o Reveals previously unknown host-virus • Limited/poor foundational knowledge of the linkage between underlying phenotypes
interactions during infection and adaptation/transmissibility provide further uncertainty
• Provides insight into the role of particular virus- • Identified host protein may not be functionally relevant or may play a minor role in
host interactions during infection viral life cycle
Alt-GoF #11 [9]: Genomic

• Proactive - can be performed using viruses that • Indirect – Identification of host proteins or virus-host interactions that contribute to
do not yet transmit between humans in nature adaptation/transmissibility provides indirect information about viral genetic and
screen to identify host
phenotypic traits underlying adaptation
factors that contribute to
fitness o Mechanistic insight may depend on prior knowledge of virus-host interactions


• Identifies host proteins that may play a role in
mammalian adaptation • Bias – Limited to investigation of previously identified phenotypic traits
Alt-GoF #12 [10]: (In vitro, o Reveals previously unknown host factors • Limited/poor foundational knowledge of the linkage between underlying phenotypes
virus-free) Proteomic or contributing to underlying phenotypes and adaptation/transmissibility provide further uncertainty
genomic screen to identify o Reveals previously unknown host-virus • Identified host protein may not be functionally relevant or may play a minor role in
host factors that interact interactions contributing to underlying viral life cycle
with particular virus phenotypes
proteins and/or contribute to
• Provides insight into the role of particular virus-
• Indirect – Identification of host proteins or virus-host interactions that contribute to
fitness adaptation/transmissibility provides indirect information about viral genetic and
host interactions phenotypic traits underlying adaptation
• Proactive - can be performed using virus gene o Mechanistic insight may depend on prior knowledge of virus-host interactions
segments from viruses that do not yet transmit
between humans in nature • Simplicity of model system – Results based on the study of a viral protein/phenotype
in isolation may not be recapitulated in the context of the full virus

Alt-GoF #13 [13]: Targeted • Limited/poor foundational knowledge of the linkage between underlying phenotypes
modification of host factor • Enables testing of the role of host markers in and adaptation/transmissibility provide further uncertainty
to alter expression or adaptation/transmissibility in the context of
function of host factors
• Identified host protein may not be functionally relevant or may play a minor role in
infection with new viral strains, to assess viral life cycle
expected to contribute to generalizability of previous findings
adaptation/transmissibility • Indirect – Identification of host proteins or virus-host interactions that contribute to
adaptation/transmissibility provides indirect information about viral genetic and
phenotypic traits underlying adaptation
• Mechanistic insight may depend on prior knowledge of virus-host interactions
• Lack of publication of negative results – Compromises evaluation of whether the
function of particular markers is broadly conserved

a GoF and alt-GoF approaches are listed in numerical order. Numbers in brackets reference the order in the landscape tables (Supplementary Information).
b Note that to date, animal-origin viruses that efficiently transmit between humans have not been discovered in nature. If available, comparative sequence and phenotypic
analysis represents a viable approach for identification of genetic markers associated with human adaptation/transmissibility
c
Blue text distinguishes an approach or outcome that is associated with GoF studies that enhance mammalian adaptation but not transmissibility. Animal passaging for the
purpose of animal model development is discussed in the Supplementary Information.

Table 15.17. Crosswalk: Use of risk mediation strategies for studies on mammalian adaptation and transmissibility
Scientific Knowledge Benefits – Utility and Limitations of Using Attenuated Strains for GoF Approaches that Enhance Mammalian Adaptation and
Transmissibility
Virusa Limitations Experimental Systemb
In vivo
High Pathogenicity • Mammalian adaptation and transmission
Strainc studies
N/A
• Animal strain In vitro
• Pathogenic reassortant • Characterization of underlying phenotypes of
mammalian adaptation and transmissibility
Genetic Context:
• Complex phenotypes are multi-genic; results may not be recapitulated in the
context of the pathogenic virus
Limited Utility:
In vivo
• Precludes study of the role of animal-origin HA and NA proteins, which are • Mammalian adaptation and transmission
Risk mediation critical viral factors in adaptation and transmissibility
studies
Reassortant-Seasonal Overlapping Phenotypes:
influenza In vitro
• Method of attenuation may alter phenotypes that contribute to • Characterization of underlying phenotypes of
adaptation/transmissibility, thus interfering with the study of
mammalian adaptation and transmissibility
Altered Course of Disease:
• Animals infected with attenuated viruses may exhibit significantly different
disease pathology, limiting the relevance to wild type viruses

Transmissibility
Limited model systems:
• Lab-adapted strains are not transmissible in ferrets
Genetic Context:
In vivo
context of the pathogenic virus • Mammalian adaptation and transmission
Risk mediation studies
Reassortant-Lab-adapted Overlapping Phenotypes:
In vitro
(e.g. PR8) • Method of attenuation may alter phenotypes that contribute to
adaptation/transmissibility, thus interfering with the study of • Characterization of underlying phenotypes of
adaptation/transmissibility mammalian adaptation and transmissibility
Altered Course of Disease:
Genetic Context:
context of the pathogenic virus In vivo
Attenuated Strain Overlapping Phenotypes: • Mammalian adaptation and transmission
• Targeted mutagenesis • Method of attenuation may alter phenotypes that contribute to studies
to remove virulence adaptation/transmissibility, thus interfering with the study of In vitro
factor (e.g. ΔMBCS) adaptation/transmissibility • Characterization of underlying phenotypes of
Altered Course of Disease: mammalian adaptation and transmissibility

Transmissibility
a
Animal-origin strains include avian- and swine-origin strains that have and have not infected humans. Pandemic strains include the 1918 H1N1, 1957 H2N2,
and 1968 H3N2 viruses. Seasonal strains include all seasonal isolates and 2009 H1N1 pandemic isolates (now circulating seasonally). Risk mediation
reassortants include all reassortants with lab-adapted viruses or with surface protein gene segments from seasonal influenza viruses. Pathogenic reassortants
include viruses with animal and/or human gene segments (both seasonal and pandemic) for which human populations have limited or no immunity.
b
The text color in the experimental system column indicates the general feasibility of the use of the virus described for in vivo or in vitro use. Green indicates
that the virus would likely be functional and representative of wild type conditions in vitro and in vivo, orange indicates that the virus may not be functional or
representative in vitro or in vivo, and red indicates that the virus would likely be non-functional in vitro or in vivo.
c
GoF approaches are shaded in blue, and alt-GoF approaches (i.e. conducting GoF approaches using attenuated strains in lieu of wild type strains) are
shaded in grey.

15.3.4 Benefits to Surveillance
Influenza pandemics occur when a novel influenza virus becomes transmissible in human populations
with limited pre-existing immunity. The likelihood and potential consequences of a pandemic are the
result of complex interactions between multiple factors related to the properties of the virus, of the host
population, and of the environment in which the virus is circulating.1407 Analysis of the phenotypic
properties of individual surveillance isolates informs the components of pandemic risk assessments
related to the properties of the virus. This section focuses on GoF benefits to these surveillance efforts.
Ultimately, GoF benefits to surveillance improve the health of human populations through public health
activities undertaken subsequent to a pandemic risk assessment, such as development of pre-pandemic
vaccines. Thus, the scope of GoF benefits to surveillance depends on the value of GoF data relative to
other factors that are considered in the risk assessment process. The process of pandemic risk assessment,
including descriptions of other factors that are considered in risk assessments as well as downstream
decision-making about pandemic preparedness policies, is described in detail in 15.3.5.
Influenza surveillance is conducted in human and animal populations, including agricultural animals,
companion animals, and wildlife. The WHO Global Influenza Surveillance and Response System
(GISRS) serves as a central repository for data about animal influenza infections in humans, generated
through passive surveillance (i.e. reporting of illnesses in patients who interact with the healthcare
system).1408 The GISRS is a two-tiered system, structured such that clinical samples from patients are
initially collected by National Influenza Centers (NICs) located throughout the world, which perform
preliminary diagnostic tests and forward samples with evidence of animal influenza infection to WHO
Collaborating Centres (WHOCCs) for thorough characterization.1409 In contrast, animal surveillance is
generally ad hoc, reflecting a combination of passive surveillance and active sampling of agricultural
animals or wildlife, and data are spread throughout several different surveillance systems. Collectively,
the goal of this surveillance is to monitor the evolution of circulating animal influenza viruses, in order to
identify those viruses that pose a risk of emerging in human populations to cause a pandemic. Resources
can then be dedicated to mitigating the risk factors associated with virus emergence, for example through
community-level interventions at the animal-human interface, and to prepare for a potential emergence
event, for example through the development of pre-pandemic vaccines.1410
Multiple virus properties contribute to the likelihood that the virus will adapt to efficiently transmit in
human populations and the potential consequences of that event:
• Whether the virus is adapted (or poised to adapt) to efficiently infect and transmit between
humans,
• Viral virulence,
• Whether the strain is antigenically similar to existing candidate vaccine viruses and stockpiled
pre-pandemic vaccines,
1408 The World Health Organization. Global Influenza Surveillance and Response System (GISRS).
http://www.who.int/influenza/gisrs_laboratory/en/. Last Update November 2, 2015. Accessed November 6, 2015.
1409 There are six WHOCCs, including the U.S. Centers for Disease Control in Atlanta, GA and St. Jude’s Children’s Research
Hospital in Memphis, TN.

• Whether the virus is sensitive to existing antivirals.
Each of these properties can be directly measured in the laboratory or can be inferred from the genetic
sequence based on the presence of molecular markers that have been linked to those phenotypes through
previous research. In practice, due to the limitations of both strategies, the strategies are utilized together.
Two other approaches are in development but are not yet used in public health practice. The first involves
the use of rapid assays to assess phenotypes underlying mammalian adaptation, transmissibility, and
virulence (i.e. versus evaluating the complex phenotype through animal experiments). The second
involves computational modeling to predict phenotype from genotype, which incorporates experimental
data about mutations that give rise to phenotypic changes, structural data, and other types of data. This
section analyzes how GoF research can improve strategies for evaluating mammalian adaptation,
transmissibility, and virulence, including strategies that are currently used and those that are in
development. The role of GoF in surveillance for antiviral resistance is evaluated in section 16.6. First,
the utility and limitations of traditional methods for laboratory evaluation of the infectivity,
transmissibility, and virulence of surveillance viruses are evaluated. This information motivates the need
for development of additional approaches that can provide information about these virus properties, the
quality of which can be improved by GoF approaches.
The pathogenicity and the ability of an animal influenza virus to infect and transmit in mammals is
typically evaluated in ferrets, though mice may also be used for pathogenicity testing. 1411 The strength of
these assays is that they directly measure the complex properties of mammalian adaptation,
transmissibility, and virulence. However, multiple shortcomings are associated with reliance on these
assays for evaluating the transmissibility and virulence of animal flu viruses collected through
surveillance. First, these assays are unable to assess when viruses have acquired underlying properties that
are necessary but not sufficient to enhance infectivity, transmissibility, or virulence, and such knowledge
about partial adaptation is of interest for pandemic risk assessments. Second, these assays require the use
of surveillance isolates, which limits the number of viruses that can be subjected to phenotypic
characterization. Although in principle, viruses can be synthetically reconstructed based on published
sequences, in practice the publicly available sequence information is often incomplete.1412 Third, viruses
may acquire mutations that alter their properties during isolation in eggs or cells, in which case the results
of the phenotypic assay will not reflect the properties of the virus present in the original clinical sample.
Fourth, transmission and virulence testing in animals requires technical expertise and must be conducted
under BSL-3 conditions, limiting the conduct of these assays to the six WHOCCs (which include the
Centers for Disease Control in Atlanta, GA and St. Jude Children’s Research Hospital in Memphis,
TN).1413,1414 Finally, when viruses of concern are initially detected abroad, political and regulatory factors
may delay the shipping of the isolate to U.S. labs, thereby delaying the generation of phenotypic data.
Although the WHO Pandemic Influenza Preparedness Framework calls for NICs to ship clinical
specimens and/or viruses that cannot be readily identified to a WHOCC or an H5 Reference Laboratory
within one week, delays arising from political and logistical factors still occur.1415,1416 U.S. select agent
regulations also considerably delay the receipt of highly pathogenic avian influenza viruses in U.S. labs.
One governmental official involved in the pandemic risk assessment process estimated that the time
1411 (2015d) Interviews with influenza researchers and government representatives involved in pandemic risk assessments.
1413 In some cases, transmissibility and virulence testing in ferrets may be conducted by university or other diagnostic labs that
have collaborative relationships with NICs.
1416 The World Health Organization. (2011b) Pandemic influenza preparedness framework for the sharing of influenza viruses
and access to vaccines and other benefits. pp. 1-67.

needed to work through regulatory logistics delays receipt of HPAI samples by three to four weeks, which
may be compounded by political or logistical issues on the part of the sending country.1417
For the reasons listed above, the CDC has incorporated the use of molecular markers for phenotypes of
concern into the pandemic risk assessment process, to complement data from animal models. Because the
phenotypes of mammalian adaptation, transmissibility, and virulence are complex, arising from the
interplay between multiple underlying phenotypes, this strategy involves inspecting sequences for
markers that are casually linked to underlying phenotypes (e.g. altered sialic acid receptor binding
specificity). Sequences may be inspected for the presence of particular mutations or for the presence of
substitutions at particular amino acid positions. In the latter case, structural analysis and molecular
modeling may be used to predict whether the substitutions has the same phenotypic effect as other,
validated substitutions at that site. Because a constellation of amino acid changes is needed for an animal
virus to evolve to efficiently infect, transmit, and cause disease in people, molecular markers are
considered collectively to determine the overall risk associated with a virus. Importantly, this process
assumes that the complex phenotypes of mammalian adaptation, transmissibility, and virulence can
accrue in a step-wise fashion, such that “partially adapted” viruses can persist in nature. (If true, the
ability to detect “partially adapted” viruses that are poised for emergence in human populations is a
strength of reliance on molecular marker data, as partial phenotypes may not be detected using phenotypic
assays for mammalian adaptation, transmissibility, and virulence.)
Influenza research experts agree that the state of this science does not enable accurate and reliable
prediction of phenotype from genotype for complex phenotypes such as mammalian adaptation,
transmissibility, and virulence. Multiple sources of scientific uncertainty limit current capabilities, which
can be broadly grouped into two categories: (1) uncertainties related to the phenotypes underlying
adaptation, transmissibility, and virulence, and (2) uncertainties related to the genetic traits that alter
underlying phenotypes.
Uncertainties related to phenotypes underlying mammalian adaptation, transmissibility, and virulence:
• Weak linkage between underlying phenotypes and adaptation/transmissibility/virulence – that is,

uncertainty in whether particular underlying phenotypes, such as altered sialic acid receptor
binding specificity, are necessary for complex phenotypes, such as mammalian adaptation across
many different virus strains.
• Lack of knowledge about how underlying phenotypes interact to alter adaptation, transmissibility,
and virulence (i.e. how to integrate the presence of multiple markers to appropriately determine
overall risk).
• Lack of knowledge about whether complex phenotypes can slowly accrue (i.e. whether partially
adapted viruses can persist in nature) or whether the acquisition of efficient infectivity,
transmissibility, and enhanced virulence in mammals is an “all-or-none” phenomenon.
Uncertainties related to the genetic traits that alter underlying phenotypes
• Inability to predict whether a particular amino acid substitution identified in one strain will have
similar phenotypic consequences in other strains.
• Lack of knowledge about whether different amino acid substitutions at a particular amino acid
position will have similar phenotypic consequences as known mutations;

• Lack of knowledge about the mutational landscape that permits evolution of a complex
phenotype – e.g. how many different sets of mutations enable the acquisition of airborne
transmissibility?
Collectively, these sources of uncertainty significantly compromise the predictive value of molecular
markers for mammalian adaptation, transmissibility, and virulence. However, the state of the science
supporting individual markers varies widely. The phenotypic consequences of certain markers, such as
the E627K mutation in the PB2 gene which lowers the optimal temperature for polymerase activity, have
been shown to be conserved in the context of multiple virus strains, and this marker has also been shown
to be enriched in human isolates of H5N1 relative to avian isolates.1418, 1419,1420,1421 (Notably, this mutation
was absent from the 2009 H1N1 pandemic virus, highlighting the point that multiple evolutionary
pathways permit adaptation of animal influenza viruses to humans.1422,1423) However, most markers are
not well-validated, either because their function is not conserved or not yet been tested across multiple
strain contexts.
Given the shortcomings associated with phenotypic assays and molecular marker data, the use of
computational methods for sequence-based predictions of phenotypes underlying mammalian adaptation,
transmissibility, and virulence has also been proposed.1424 Although a variety of computational methods
have shown promise for predicting phenotype from genotype, for those “known” phenotypes associated
with adaptation/transmissibility, the accuracy of their predictions remains largely unknown.1425,1426
GoF approaches have potential to address shortcomings associated with the use of virological data,
molecular markers, and computational methods to evaluate the infectivity, transmissibility, and virulence
of animal influenza viruses in mammals, representing three different strategies for improving upon the
status quo. The value of each strategy and the utility and limitations of GoF approaches for improving
each strategy, relative to alt-GoF approaches, are discussed below.
15.3.4.1 Analysis of GoF and alt-GoF approaches that support the development of rapid phenotypic
assays
GoF approaches that identify new phenotypes underlying mammalian adaptation, transmissibility, and
virulence, that strengthen the linkage between underlying phenotypes and complex phenotypes, and that
provide insight into how underlying phenotypes synergize to alter host tropism, transmissibility, and
virulence provide a foundation for the development of rapid phenotypic assays.
1418 Qi L et al (2014) Contemporary Avian Influenza A Virus Subtype H1, H6, H7, H10, and H15 Hemagglutinin Genes Encode
a Mammalian Virulence Factor Similar to the 1918 Pandemic Virus H1 Hemagglutinin. mBio 5: e02116-02114
1419 Steel J et al (2009) Transmission of Influenza Virus in a Mammalian Host Is Increased by PB2 Amino Acids 627K or
627E/701N. PLoS pathogens 5: e1000252
1420 Le QM et al (2009) Selection of H5N1 influenza virus PB2 during replication in humans. Journal of virology 83: 5278-5281
1421 Luk GS et al (2015) Transmission of H7N9 Influenza Viruses with a Polymorphism at PB2 Residue 627 in Chickens and
Ferrets. Ibid. 89: 9939-9951
1422 Bussey KA et al (2010) PB2 residue 271 plays a key role in enhanced polymerase activity of influenza A viruses in
mammalian host cells. Ibid. 84: 4395-4406
1423 Herfst S et al ibid.Introduction of virulence markers in PB2 of pandemic swine-origin influenza virus does not result in
enhanced virulence or transmission. 3752-3758

15.3.4.1.1 Strengths and weaknesses of using rapid phenotypic assays to inform pandemic risk
assessments
Rapid assays to measure phenotypes underlying mammalian adaptation, transmissibility, and virulence
could be performed in lieu of traditional evaluation of these complex phenotypes using ferrets. The
development of rapid phenotypic assays holds promise for improving analysis of surveillance data for
several reasons. First, the use of assays that are higher throughput than ferret testing will enable the
phenotypic characterization of a larger number of viruses. Second, for those assays interrogating the
function of a single protein or a protein complex, synthesizing the relevant genes based on publicly
available genetic sequence data may be feasible, which would enable the characterization of viruses for
which isolates are not available. In the event that in vitro, virus-free rapid phenotypic assays can be
developed, these assays would pose lower lab safety risks than ferret testing using full, infectious virus.
Third, rapid phenotypic assays that require less technical expertise than ferret experiments are better
suited for NICs, which would shorten the time lag between the initial detection and phenotypic
characterization of a given virus. Thus, taken together, the development of rapid phenotypic assays has
the potential to expand the quantity and the timeliness of phenotypic characterization data available for
pandemic risk assessments. However, these assays will need to be carried out under BSL-3 conditions,
which will limit the number of diagnostic laboratories that will be able to conduct the assays. Notably, the
majority of NICs do not have BSL-3 capabilities, particularly in countries in which animal influenza
viruses of concern are circulating (as BSL-3 capabilities are not needed for isolation of seasonal influenza
viruses, which comprises the bulk of the diagnostic workload of NICs). That said, the number of NICs
with BSL-3 capabilities themselves (or with access to BSL-3 labs through collaborative relationships with
university labs, U.S. military labs such as NAMRU-3, or other labs) has increased since 2005 and is likely
to continue to increase.1427,1428 Though challenging due to the expense and technical expertise needed to
construct and run a BSL-3 lab, this increase will facilitate the conduct of rapid phenotypic assays using
whole viruses in the future.
across many strain contexts. Additionally, interpretation of the results requires knowledge about how
the phenotypes underlying mammalian adaptation, transmissibility, and virulence, described above,
constrain the development and use of rapid phenotypic assays. As discussed in detail in sections 15.3.3.3
and 15.4.3.1, both GoF and alt-GoF approaches can provide insight into these scientific questions. The
relevant findings are summarized below.
15.3.4.1.2 Summary – benefits of GoF approaches relative to alt-GoF approaches
GoF approaches represent the most efficient and effective approach for identifying novel phenotypic traits
underlying mammalian adaptation, transmissibility, and virulence. Critically, GoF approaches are
uniquely capable of discovering phenotypic traits underlying the transmissibility of animal influenza
viruses because these viruses do not efficiently transmit between humans in nature. Furthermore, targeted
genetic modification of viruses to introduce genetic traits that alter underlying phenotypes is uniquely
capable of demonstrating that a particular phenotype is causally linked to enhanced
infectivity/transmissibility/ virulence in mammals across multiple virus contexts. Additionally, the ability
to alter phenotypes individually and in combination (i.e. through incorporation of varying sets of
1428 Navy Medical Research Center. Naval Medical Research Unit 3 (NAMRU-3) Cairo, Egypt.
http://www.med.navy.mil/sites/nmrc/Pages/namru3.htm. Last Update Accessed November 28, 2015.

mutations) provides insight into how multiple underlying phenotypes interact to enhance infectivity,
transmissibility, or virulence in mammals. This approach can also determine how an “intermediate” level
of adaptation/transmissibility/ virulence, i.e. acquisition of some but not all phenotypic traits that are
required for viruses to efficiently infect, cause disease, and transmit in mammals, affects viral fitness,
which may provide insight into whether such partially adapted strains can persist in nature. However, the
major caveat associated with GoF approaches is that results gleaned from laboratory studies involving
animal models may not translate to human disease in nature.
Alternative approaches have significant limitations relative to GoF approaches. Characterization of wild
type viruses provides limited insight into phenotypic traits underlying mammalian adaptation and
transmissibility because animal influenza viruses that efficiently infect and transmit in humans do not
exist in nature. However, characterizing the constellation of underlying phenotypes present in a large
number of wild type viruses (e.g. sialic acid receptor binding specificity, HA stability, optimal
temperature for polymerase activity, etc.) is uniquely capable of providing insight into whether viruses
that have a subset of the properties that are necessary for enhanced infectivity, transmissibility, or
virulence can persist in nature.
LoF approaches have limited utility for broad and unbiased identification of phenotypic traits that
contribute to transmissibility and pathogenicity due to their inefficiency, as a limited number of mutants
can been screened through ferret transmission studies due to technical and ethical concerns and mutants
may recover transmissibility during the single round of infection needed for characterization. An
additional limitation is that the fact that mechanisms underlying transmissibility of seasonal/pandemic
viruses may not translate to animal influenza viruses. Though LoF approaches can be used to causally
demonstrate that a particular phenotype is necessary for efficient transmissibility and enhanced virulence,
this approach cannot be used to understand how multiple phenotypes synergize to enhance infectivity,
transmissibility, or virulence. This information critically informs how results from multiple phenotypic
assays should be integrated to evaluate overall pandemic potential. Surveillance-based approaches,
including comparison of human and animal isolates, comparison of sequences spanning avian to
mammalian adaptation events, and comparison of viral isolates with varying levels of virulence are
limited to the study of previously known traits and provide associative data. Notable exceptions include
the analysis of precursor/spillover pairs, for the study of adaptation/transmissibility, and analysis of viral
isolates over the course of infection in a single patient, for the study of virulence. However, the
availability of both types of paired isolates is low. In addition, neither surveillance-based approaches nor
LoF approaches can provide insight into phenotypes underlying transmissibility because animal influenza
viruses that efficiently transmit in humans do not exist in nature. In vitro, virus free approaches, which
involve the study of known phenotypes in isolation, cannot provide information about the functional
relationships among underlying phenotypes or between underlying phenotypes and
adaptation/transmissibility.
15.3.4.2 Analysis of GoF and alt-GoF approaches that support the use of molecular markers to
evaluate the risk posed by circulating animal influenza viruses
GoF approaches support the use of molecular marker data to evaluate the risk posed by circulating animal
influenza viruses in two ways: (1) through the discovery of novel molecular markers of phenotypic
properties of concern, and (2) by strengthening the predictive value of known molecular markers.
15.3.4.2.1 Strengths and weaknesses of using molecular marker data to inform pandemic risk assessments
The use of molecular marker data to evaluate the pandemic potential of animal influenza viruses has
several strengths relative to the use of phenotypic data. In particular, the fact that clinical isolates can be
directly sequenced provides several advantages. First, direct sequencing of clinical isolates avoids the

problem that the composition and properties of viral species present in the clinical sample could change
during the virus isolation process. Second, following inactivation of virus present in a clinical sample, the
sequencing procedure can be carried out under BSL-2 conditions and thus can more feasibly be
implemented at NICs and other diagnostic labs in developing countries. Third, whether from clinical
samples or virus isolates, sequencing is becoming ever cheaper and easier. As a result, viral genetic
sequence data is currently the fastest and most reliable data generated by diagnostic labs in areas where
viruses of concern are circulating.1429 However, most genetic surveillance data is generated by sequencing
the HA and NA genes of viral isolates at WHOCCs.1430 Full realization of the benefits that can be derived
from the use of molecular marker data will require expanding the sequencing capabilities of diagnostic
laboratories that comprise the “base” of the influenza surveillance system as well as increasing the
proportion of clinical samples that are directly sequenced. Additionally, the number of viruses that are
subjected to whole genome sequencing (as opposed to sequencing the HA and NA genes only) must be
increased in order to fully utilize molecular markers in genes other than HA and NA. Notably,
stakeholders throughout the surveillance system recognize that capabilities in each of these areas –
sequencing at NICs, direct sequencing of clinical samples, and whole genome sequencing – are desirable
and are striving to implement them whenever and wherever possible.1431
As described above, the current utility of molecular markers to the interpretation of genetic surveillance
data is constrained by multiple sources of scientific uncertainty. Additionally, as knowledge about the
phenotypes underlying mammalian adaptation, transmissibility, and virulence is incomplete, the
discovery of additional molecular markers associated with novel underlying phenotypes would broaden
the utility of this approach. As discussed in detail in section 15.3.3.3 and 15.4.3.1, both GoF and alt-GoF
approaches can provide insight into these scientific questions. The relevant findings are summarized
below.
The benefits of GoF approaches relative to alt-GoF approaches for addressing knowledge gaps at the
phenotypic level were summarized in section 15.3.4.1.2. In brief, GoF approaches represent the most
efficient and effective methods for discovering novel phenotypes underlying adaptation/transmissibility/
virulence and are uniquely capable of demonstrating that phenotypes are causally linked to enhanced
infectivity/transmissibility/virulence of animal influenza viruses in representative animal models. GoF
approaches are also uniquely capable of providing definitive information about how multiple phenotypes
synergize to promote mammalian adaptation, efficient transmissibility, and virulence. However, alt-GoF
approaches, namely characterization of wild type viruses, are uniquely capable of demonstrating whether
partially adapted viruses exist in nature, which provides insight into whether complex phenotypes such as
adaptation, transmissibility, and virulence can accrue in a step-wise fashion (an underlying assumption of
the use of molecular markers to evaluate pandemic risk).
Both GoF and alt-GoF approaches can provide insight into the scientific knowledge gaps related to the
genetic traits underlying mammalian adaptation, transmissibility, and virulence. GoF approaches
represent the most efficient and effective approach for identifying novel genetic traits underlying
mammalian adaptation, transmissibility, and virulence. Furthermore, targeted genetic modification of
viruses to introduce genetic traits associated with mammalian adaptation/transmissibility/virulence is
uniquely capable of demonstrating that particular genetic markers are necessary and sufficient for
mammalian adaptation, transmissibility, or enhanced virulence across multiple virus contexts. In addition,
GoF approaches, namely forward genetic screens, are uniquely capable of systematically exploring
1430 Ibid.
1431 Ibid.

alternative mutational pathways for altering an underlying phenotype (e.g. changing sialic acid receptor
binding specificity) in the context of whole virus. The major caveat associated with GoF approaches is
that results gleaned from laboratory studies involving animal models may not translate to the potential to
cause a pandemic in humans.
Alternative approaches have significant limitations relative to GoF approaches. Characterization of wild
type viruses provides limited insight into genetic traits underlying mammalian adaptation and
transmissibility because animal influenza viruses that efficiently infect and transmit in humans do not
exist in nature. LoF approaches have limited utility for broad and unbiased identification of novel genetic
traits that are necessary for transmissibility or enhanced virulence due to their inefficiency and the fact
that mechanisms underlying transmissibility of seasonal/pandemic viruses may not translate to animal
influenza viruses. Surveillance-based approaches, including comparison of human and animal isolates and
of sequences spanning avian to mammalian adaptation events, have limited utility for the discovery of
novel genetic traits associated with adaptation/transmissibility/virulence due to the high genetic diversity
of influenza viruses and shortcomings in the quality and availability of surveillance data. A notable
exception is the comparison of genetically similar viruses such as precursor/spillover strains and the
comparison of viral isolates over the course of illness in a single patient, though such paired isolates are
rarely available. However, surveillance-based approaches have several unique strengths for validating the
functional consequences of particular markers. Comparison of human and animal isolates or of human
isolates with varying levels of virulence is uniquely capable of providing direct insight into traits
associated with human adaptation and virulence across multiple strain contexts. These traits can be
considered “causally” linked if a large enough number of sequences are compared. Notably, this approach
cannot be used to validate markers associated with enhanced transmissibility because animal influenza
strains that transmit efficiently between humans do not exist in nature. The high-throughput nature of in
vitro, virus free approaches relative to animal experiments renders them appealing for the discovery of
additional mutations that give rise to particular phenotypic changes (through forward genetic screens) and
for validating the function of particular markers in new genetic contexts. However, results may not be
recapitulated in vivo, in the context of the full virus.
Notably, the feasibility of using molecular markers to infer phenotype from genotype depends on several
factors: (1) the extent to which the functional consequences of particular markers are conserved across
multiple strain contexts, (2) the number of different sets of mutations that give rise to a phenotype of
interest, and (3) whether the phenotypic changes associated with adaptation/transmissibility/virulence
arise due to the concerted effects of many mutations, each of which has a small individual effect, or
whether single mutations give rise to large phenotypic changes. Influenza researchers emphasized that for
the known phenotypes associated with adaptation, transmissibility, and virulence, the answers to these
questions are unknown and are likely to vary by phenotype. For example, it is likely that a large number
of distinct mutations are capable of increasing HA stability and that the set of mutations that increase HA
stability will vary by strain. Thus, this phenotype may not be a good candidate for the molecular marker
approach, but rather for the rapid phenotypic assay approach. Several researchers felt that performing a
limited number of GoF experiments to address each of these three questions would enable the
determination of whether delineating the set of mutations that can give rise to a particular phenotype is
achievable through a reasonable number of experiments.
15.3.4.3 Analysis of GoF and alt-GoF approaches that improve predictive models
GoF experiments that provide data about whether particular mutations alter phenotypes of concern have
potential to improve existing computation models for predicting phenotype from genotype.

15.3.4.3.1 Strengths and weaknesses of using computational models to inform pandemic risk assessments
As the use of computational models to predict phenotypes underlying mammalian adaptation,

transmissibility, and virulence capitalizes on (and depends on) the availability of sequence data, the
strengths and limitations of this approach relative to the use of virologic data are similar to those
described above for the use of molecular marker data.
Existing computational models cannot reliably predict phenotypes underlying mammalian adaptation,
transmissibility, and virulence based on sequence information. Additional experimental data is needed to
appropriately parameterize models, and experiments must be conducted to validate the phenotypic
predictions of models. Both GoF and alt-GoF approaches can generate data that improves the accuracy of
existing models.
15.3.4.3.2 Summary - benefits of GoF approaches relative to alt-GoF approaches
about mutations that do and do not give rise to phenotypic changes of interest. These data are critical for
building models that can account for the context dependence of genetic changes in influenza biology. GoF
approaches (targeted mutagenesis and forward genetic screens) are uniquely capable of generating these
data in the context of the full virus, although in vitro, virus free approaches can also be used.
In contrast, additional experimental data about the biophysical basis of underlying phenotypes, such as
crystallography data and measurements of HA binding affinities to α2,6 and α2,3 sialoglycans, is also
needed to improve existing models. These data are generated through alternative experimental
approaches.
Finally, model predictions must be validated experimentally, and results feedback to improve model
accuracy. While predictions can be tested using in vitro, virus free assays, experimental validation in the
context of the full virus (GoF) is also important.
Taken together, GoF and alt-GoF approaches provide different types of experimental data that are both
essential for improving the accuracy of predictive models, and GoF approaches are uniquely capable of
validating model predictions in the context of the full virus.
15.3.4.4 Summary – Benefits of GoF approaches relative to alt-GoF approaches to surveillance
A key goal of influenza surveillance is to monitor the evolution of circulating animal influenza viruses, in
order to identify those viruses that pose a risk of emerging in human populations to cause a pandemic.
Resources can then be dedicated to mitigating the risks of an emergence event. Analysis of the phenotypic
properties of individual surveillance isolates is an important aspect of pandemic risk assessments,
including transmissibility and virulence in mammals. Currently, this analysis relies on the laboratory
characterization of surveillance isolates and, to a lesser extent, the inspection of sequences for molecular
markers associated with phenotypes underlying mammalian adaptation, transmissibility and virulence.
Both methods exhibit shortcomings that compromise the accuracy, timeliness, and quantity of data. Two
additional approaches are in development to address these shortcomings: rapid assays for phenotypes
underlying mammalian adaptation and transmissibility, and computational models to predict underlying
phenotypes from genotype. Such rapid phenotypic assays do not yet exist, and the prospective accuracy of
existing models is unknown. Both GoF and alt-GoF experimental approaches have potential to address
shortcomings associated with the use of rapid phenotypic assays, molecular markers, and computational
models.

GoF approaches provide unique benefits to the design and validation of rapid assays for phenotypes
underlying adaptation, transmissibility, and virulence. The fact that these assays would be high-
throughput, less technically challenging than ferret experiments, and could likely utilize synthetically
generated viral gene segments could increase the quantity and timeliness of phenotypic data available,
relative to the use of traditional phenotypic characterization assays for adaptation, transmissibility, and
virulence. The accuracy and utility of rapid phenotypic assays depends on establishing a strong linkage
between underlying phenotypes and adaptation/transmissibility/virulence as well as developing an
understanding of how multiple phenotypes synergize to enhance the infectivity, transmissibility, and
virulence of animal influenza viruses in mammals. GoF approaches represent the most efficient and
effective approach for discovering novel phenotypes underlying mammalian adaptation, transmissibility,
and virulence and are uniquely capable of demonstrating that a particular phenotype is causally linked to
enhanced infectivity/transmissibility/virulence in mammals across multiple virus contexts. GoF
approaches are also uniquely capable of causally determining how multiple underlying phenotypes
interact to enhance infectivity, transmissibility, or virulence in mammals, which provides insight into how
information about underlying phenotypes should be integrated for a risk assessment. However, a major
caveat associated with GoF approaches is that results gleaned from laboratory studies involving animal
models may not translate to human disease in nature. Characterizing the constellation of underlying
phenotypes present in a large number of wild type viruses (alt-GoF) is uniquely capable of providing
insight into whether partially adapted viruses can persist in nature, which lends support to the practice of
inferring complex phenotypes such as adaptation, transmissibility, and virulence based on data about
underlying phenotypes. Ultimately, the utility of these assays depends on whether phenotypes underlying
mammalian adaptation, transmissibility, and virulence are conserved across different strains, which is not
yet well-understood. However, the fact that the same underlying phenotypes were shown to confer
airborne transmissibility to two very different H5N1 strains – a fully avian clade 2.1 H5N1 strain and an
H5N1 reassortant strain containing an avian clade 1 H5 gene and the remaining genes from a 2009 H1N1
pandemic virus – suggests that conserved mechanisms may exist.1432,1433 Finally, a notable barrier to
realization of the benefits derived from the use of rapid phenotypic assays is that these assays must be
carried out under BSL-3 conditions, which limits the number of diagnostic laboratories that will be able to
conduct the assays. Most NICs do not have BSL-3 capabilities, though the number of NICs with BSL-3
labs is increasing.1434
GoF approaches provide unique benefits to the practice of using molecular markers to infer phenotypes
underlying adaptation/transmissibility/virulence based on genetic sequence data. As sequencing has
become cheaper and easier, sequence data has become the fastest and most reliable type of surveillance
data produced by diagnostic labs located in countries in which animal influenza viruses of concern are
circulating. Furthermore, the increasing reliance on direct sequencing of clinical samples has potential to
increase the accuracy of phenotypic characterization information, relative to sole reliance on traditional
phenotypic assays using viral isolates. Currently, most molecular markers for mammalian adaptation,
transmissibility, and virulence have low predictive value due to significant scientific uncertainties
regarding the association between underlying phenotypes and adaptation/transmissibility/virulence,
whether the function of markers is conserved across different strain contexts, and the breadth of mutations
that can give rise to a particular phenotypic change. Additionally, it is likely that as-yet-undiscovered
genetic and phenotypic traits contribute to mammalian adaptation, transmissibility, and virulence. As
discussed above, GoF approaches provide essential data for strengthening the linkage between underlying
phenotypes and adaptation/transmissibility/virulence. GoF approaches also provide unique advantages for
discovering novel markers and strengthening the predictive value of known markers. Namely, GoF

approaches represent the most efficient and effective approach for discovering novel genetic traits
underlying mammalian adaptation, transmissibility, and virulence and are uniquely capable of
demonstrating that particular genetic markers are necessary and sufficient for mammalian adaptation,
transmissibility, or virulence across multiple virus contexts. However, the validation of molecular
markers for mammalian adaptation or virulence through analysis of genetic surveillance data (alt-GoF) is
uniquely capable of providing direct insight into traits associated with human adaptation/virulence across
multiple strain contexts, which complements GoF approaches. Notably, surveillance-based approaches
are not viable for the validation of molecular markers associated with transmissibility because animal
influenza strains that transmit efficiently between humans in nature do not exist. GoF approaches are also
uniquely capable of systematically exploring alternative mutational pathways for modifying an
underlying phenotype in the context of whole virus. In vitro, virus free approaches can also be used, but
results may not be recapitulated in the context of the full virus. As above, the major caveat associated
with GoF approaches is that results gleaned from laboratory studies involving animal models may not
translate to human disease in nature. Critically, the feasibility of using molecular markers to infer
phenotype from genotype will depend on the functional generalizability of particular markers, the breadth
of the mutational landscape for a particular phenotypic change, and the extent to which individual
mutations alter a particular phenotype. The answers to these questions are unknown and are likely to vary
by phenotype. A limited number of GoF experiments will enable researchers to determine whether
delineating the set of mutations that can give rise to a particular phenotype is achievable through a
reasonable number of experiments. Finally, molecular markers that confer large phenotypic changes are
much more useful than molecular markers that minimally modify a phenotype of interest, as integrating
many mutations that each have small individual effects is likely to be difficult. To date, some markers
have been found to confer substantial phenotypic changes while others have minor effects, and future
discoveries are likely to be similarly mixed. Finally, notable barriers to the full realization of benefits
derived from the use of molecular markers include the need to further increase the number of sequences
generated at NICs, the number of clinical samples that are directly sequenced, and the number of viruses
that are subjected to whole genome sequencing.
GoF approaches are also critical for improving models for prediction of underlying phenotypes based on
sequence data. Specifically, GoF approaches that generate information about mutations that do and do not
give rise to phenotypic changes of interest provide critical training data for models, and GoF approaches
are needed to validate model predictions in the context of the full virus. Importantly, other types of
biophysical data generated through alternative experimental approaches are also critical for improving the
accuracy of existing models. In addition to scientific advancements, full realization of the benefits derived
from the use of computational models will require expanding the sequencing capabilities of influenza
surveillance networks as described above.
The utility and limitations of different approaches for evaluating the transmissibility and virulence of
circulating animal influenza viruses are summarized in Table 15.18, below. Both the direct measurement
of virus phenotypes in the laboratory and the prediction of underlying phenotypes from genotype, either
through sequence inspection for molecular markers or computational modeling approaches, have inherent
strengths and limitations. Namely, the generation of phenotypic data will always be delayed by the need
to ship clinical samples or viral isolates, and viruses may acquire adaptive changes that alter their
phenotypic properties during isolation. However, direct measurements of phenotypic properties are
invaluable. In contrast, as sequence data is increasingly available from NICs and other “base” level
diagnostic laboratories, the application of predictive methods will enable the rapid generation of
phenotypic “data” that reflects the properties of viruses present in clinical samples, allowing for more
rapid characterization of emerging influenza viruses. However, due to the inherent uncertainties
associated with predictions, the subsequent confirmation of predictions through phenotypic testing is
critical. Therefore, virological data and sequence-based predictive data are complementary, and

consideration of both will strengthen the timeliness and accuracy of assessments of virus properties that
contribute to pandemic risk.

Table 15.18. Summary of the Benefits of GoF Approaches that Enhance Transmission and Virulence in Mammals
Surveillance benefits – Aid evaluation of the transmissibility and virulence of circulating animal influenza viruses
• Provides a direct readout of phenotypes underlying mammalian • Reliable rapid phenotypic assays do not yet
adaptation, transmissibility, and virulence exist, and their future validity depends on
• Could expand the quantity of phenotypic characterization data scientific advancements
available: o Timeframe for establishing that knowledge is
GoF #1: o High-throughput assays will enable the characterization of a large uncertain, likely to be long-term
Support the development of number of surveillance isolates • Broad utility of rapid phenotypic assays will
rapid assays for phenotypes • Could increase the timeliness of phenotypic characterization data depend on whether mechanisms underlying
underlying mammalian available: mammalian adaptation, transmissibility, and
adaptation, transmissibility, o Relatively simple execution of rapid phenotypic assays relative to virulence are conserved across different strains
and virulence ferret testing experiments will enable testing at NICs, abrogating o Not yet well-understood
the need to ship samples to WHOCCs for characterization • The need to conduct assays involving whole
• Enables detection of viruses that are “partially adapted” virus under BSL-3 conditions will limit the
o Viruses that exhibit changes in one or more underlying number of diagnostic labs that can carry out
phenotypes these assays

• Could increase the accuracy of phenotypic characterization data:
o Clinical samples can be directly sequenced
• Could increase the timeliness of phenotypic characterization data: • Significant scientific uncertainties compromise
the current utility of molecular markers for
o NICs and other field diagnostic labs are increasingly capable of
mammalian adaptation, transmissibility, and
sequencing virus samples, abrogating the need to ship samples to
virulence
WHOCCs for characterization
GoF #2: o Time frame for establishing that knowledge is
• Could expand the quantity of phenotypic characterization data: uncertain, likely to be long-term
Strengthen the predictive
o As sequencing becomes cheaper and easier, whole genome
value of molecular markers
sequencing of viruses collected through surveillance will become
• Use of molecular markers is inherently
for mammalian adaptation, predictive
transmissibility, and increasingly common
virulence • Enables detection of viruses that are “partially adapted” • Full realization of benefits depends on
expanding sequencing capabilities at NICs, as
o Viruses that exhibit changes in one or more underlying well as increasing the number of viruses that are
phenotypes subjected to whole genome sequencing and the
• Molecular marker data are currently used to interpret surveillance number of clinical samples that are directly
data sequenced
o New data can be incorporated into the process in the immediate
term

• Increase the accuracy of phenotypic characterization data: • Reliable computational models for phenotypes
underlying mammalian adaptation,
o Clinical samples can be directly sequenced transmissibility, and virulence do not yet exist,
• Increase the timeliness of phenotypic characterization data: and their future validity depends on scientific
GoF #3: advancements
o NICs and other field diagnostic labs are increasingly capable of
Support development of sequencing virus samples, abrogating the need to ship samples to o Timeframe for establishing that knowledge is
computational models for WHOCCs for characterization uncertain, likely to be long-term
predicting phenotypes
underlying mammalian • Increase the quantity of phenotypic characterization data: • Use of computational models is inherently
adaptation, transmissibility, o As sequencing becomes cheaper and easier over time, whole predictive
and virulence based on genome sequencing of viruses collected through surveillance will • Full realization of benefits depends on
sequence become increasingly common expanding sequencing capabilities at NICs, as
• Enables detection of viruses that are “partially adapted” well as increasing the number of viruses that are
subjected to whole genome sequencing and the
o Viruses that exhibit changes in one or more underlying number of clinical samples that are directly
phenotypes sequenced
• Assays are unable to detect when viruses have

acquired underlying phenotypic changes that are
necessary but not sufficient to alter infectivity,
transmissibility, and virulence in mammals (i.e.
“partially adapted” viruses)
Alt-GoF #1:
• The number of viruses that can be characterized
Phenotypic evaluation of is limited by the availability of surveillance
mammalian adaptation, • Provides direct readout of infectivity, transmissibility, and virulence isolates
transmissibility, and in appropriate animal models
virulence in ferrets or other • Sample shipping delays due to political and
appropriate animal models regulatory factors delay the generation of
phenotypic data
o Due to the technical expertise and
biocontainment conditions required for these
assays, they are currently conducted at
WHOCCs only

15.3.5 Benefits to decision-making in public health policy
GoF approaches that enhance the infectivity and transmissibility of animal influenza viruses in
representative animal models have potential to benefit pandemic preparedness planning in two ways.
First, the demonstration that avian influenza viruses can evolve the capacity for more efficient
transmission in mammals may, in and of itself, stimulate interest and investment in pandemic
preparedness initiatives. The second benefit derives from GoF benefits to surveillance. Analysis of the
phenotypic properties of animal influenza surveillance isolates plays a critical role in assessment of their
pandemic risk, as described in detail below. In turn, pandemic risk assessments inform decision-making
about how to invest in public health preparedness activities for influenza pandemics. Thus, GoF-derived
improvements to the analysis of influenza surveillance data could have downstream benefits to decision-
making in public health policy. This section evaluates the potential benefits of each type of GoF data,
relative to alternative approaches, in turn.
15.3.5.1 Benefits of “proof of principle” GoF research that demonstrates the capacity of a virus to
evolve more efficient transmissibility in representative animal models
Researchers have suggested that the “proof of principle” demonstration that an animal influenza virus can
evolve the capacity for airborne transmission in a laboratory setting, as a blunt indicator of the pandemic
potential of the virus, could inform government interest and investment in pandemic preparedness
initiatives. However, pandemic preparedness activities at the U.S. CDC and ASPR, including BARDA,
did not change in the wake of the 2012 demonstration that H5N1 could evolve the ability to transmit via
the airborne route between ferrets, suggesting that this is not a real benefit.1435,1436,1437 CDC and BARDA
representatives noted that the level of resources dedicated to H5N1 preparedness was already high at the
time those papers were published, as many CVVs had been developed and a quantity of pre-pandemic
vaccine doses had been developed stockpiled.1438 Thus, there may have been minimal room for increasing
the level of USG investment in preparedness for that virus sub-type. However, pandemic preparedness
activities also did not change in response to the laboratory demonstration that avian influenza H9N2 could
acquire the capacity for airborne transmission in ferrets, which provides an instructive comparison.1439 At
that time, multiple CVVs for H9N2 had been developed, but decision-makers had chosen not to proceed
further along the pre-pandemic vaccine production pipeline because H9N2 had caused fewer and milder
cases than H5N1.1440 This finding indicates that the epidemiological differences between H5N1 and H9N2
human infections were responsible for the initial differences in the level of resources dedicated to
preparedness for each virus. However, the laboratory transmission results did not change this decision,
suggesting that for viruses that have already caused human infections, additional laboratory data will not
significantly influence decision-making related to pandemic preparedness.1441 USG representatives
involved in pandemic preparedness indicated that the response to the demonstration that an animal virus
that has not yet caused human infections can evolve the capacity for airborne transmission would also be
minimal, due to the lack of certainty about whether laboratory results translate to humans in nature.1442 If
the virus were known or suspected to be circulating in animal populations in the US, enhanced
1437 (2015i) Interviews with CDC, ASPR, and BARDA representatives.
1439 Sorrell EM et al (2009) Minimal molecular constraints for respiratory droplet transmission of an avian-human H9N2
influenza A virus. Proc Natl Acad Sci U S A 106: 7565-7570
1441 Ibid.
1442 Ibid.

surveillance might be undertaken to better understand the prevalence and geographic distribution of the
virus in nature. However, the result would be highly unlikely to trigger investments in pre-pandemic
vaccine development. Notably, this result may impact pandemic preparedness planning in developing
countries in which high-risk viruses are circulating, as discussed in the “globalization of benefits” section
(15.9).
15.3.5.2 Benefits of GoF research that informs pandemic risk assessments
The second mechanism through which GoF approaches can benefit pandemic preparedness planning is
through pandemic risk assessments, downstream of GoF benefits to surveillance. As discussed in section
15.3.4, GoF approaches have potential to benefit virological surveillance (i.e. by supporting the
development of rapid phenotypic assays) as well as genetic surveillance (i.e. by strengthening the
predictive value of molecular markers for phenotypic properties of concern and by improving
computational models for predicting phenotype from genotype). The use of molecular markers for
phenotypic properties of concern is currently incorporated into the risk assessment process, as described
in detail below. As neither rapid assays nor robust computational models for relevant phenotypes exist,
how results from notional future assays/models would be considered in risk assessments is uncertain.
Thus, the potential benefits of rapid phenotypic assays or computational models to pandemic risk
assessments is not formally evaluated in this section, but a discussion of how results from either could
contribute to the risk assessment process is provided at the end of the section.
This section analyzes the value of using molecular marker data relative to other types of data that are
considered in the pandemic risk assessment process (i.e. epidemiological and ecological data), which
provides an “upper bound” to the public health benefits that can be achieved through GoF improvements
to surveillance. First, to provide context for this analysis, current strategies for pandemic risk assessments
are reviewed, and shortcomings in existing processes are highlighted.
15.3.5.2.1 Background – pandemic risk assessment and strategies for decision-making about investments
in pandemic preparedness
Influenza pandemics occur when a novel influenza virus becomes transmissible in human populations
with limited or no pre-existing immunity. Due to the complex interplay between virus, host, and
ecological factors that shape viral evolution in nature, predicting the timing of the next influenza
pandemic and the strain that causes it is not possible.1443 Nonetheless, given the high public health burden
associated with annual influenza epidemics and past influenza pandemics (see chapter 5), the US
government undertakes influenza pandemic preparedness activities to bolster US capabilities for rapid
detection of novel influenza events and to limit the spread of disease, death, and potential societal impacts
if/when the next influenza pandemic occurs.1444 Some preparedness efforts target particular influenza
strains or sub-types, including the development of novel diagnostics, enhanced animal or public health
surveillance, and the development of pre-pandemic vaccines, while others are largely strain-agnostic,
such as stockpiling antivirals. In particular, the development of pre-pandemic vaccines is a key aspect of
pandemic preparedness because influenza vaccination is the primary public health strategy for reducing
influenza-associated morbidity and mortality during outbreaks.1445
1444 Ibid.
1445 Ampofo WK et al (2013b) Strengthening the influenza vaccine virus selection and development process: outcome of the 2nd

As resources for pandemic preparedness efforts are limited, a major challenge is determining how
resources for strain-specific investments should be allocated, in particular for the development of pre-
pandemic vaccines. The U.S. National Strategy for Pandemic Influenza (2005) calls for a risk- and
evidence-based approach to guide comprehensive planning and response efforts.1446 To that end, the CDC,
in collaboration with subject matter experts in influenza virology, diagnosis, epidemiology, ecology, and
laboratory research in animal and human influenza, developed a framework for assessing the relative risk
posed by emerging influenza viruses and an accompanying tool – the Influenza Risk Assessment Tool
(IRAT). Those results then inform prioritization of resources for preparedness efforts directed at
particular strains/sub-types.
The IRAT provides a formal method for evaluating the relative risk posed by different emerging influenza
strains (e.g. H5N1 versus H7N9).1447,1448 This method is based on subject matter expert input about risk
elements that govern the likelihood that a particular strain will adapt to efficiently transmit in human
populations and the expected public health consequences of that emergence event. These risk elements
can be broadly grouped into four categories:
• Elements relating to the properties of the virus (e.g. transmissibility and virulence),
• Elements relating to the attributes of host populations (e.g. the degree of pre-existing immunity),
• Elements relating to epidemiology, and
• Elements relating to ecological factors (e.g. the extent of human infections and the prevalence and
geographic distribution of the virus in animal populations).
Selected elements will be described in more detail below. Risk elements pertaining to the properties of the
virus are informed by virological data (e.g. transmission studies in ferrets) and by genomic data, including
molecular marker data (e.g. whether molecular markers associated with enhanced transmissibility in
ferrets are present in the viral genetic sequence). Individual risk elements have been weighted, based on
SME input about their relative contribution to the likelihood and expected consequences of emergence of
particular strains, and all elements are considered collectively to determine an overall risk score. Notably,
the relative weighting factors for distinct risk elements are different for the “likelihood of emergence” and
“consequences” parts of the tool.
Only some emerging viruses are subjected to formal risk assessments using the IRAT, and not all
pandemic preparedness decisions related to those viruses are based on formal risk assessment scores.
However, the risk elements outlined in the IRAT are considered when informally evaluating risks posed
by emerging influenza viruses. Thus, the following analysis of how GoF benefits to surveillance could
improve the pandemic risk assessment process and downstream decision-making represents the value of
GoF insights to decision-making about preparedness for emerging influenza outbreaks in general.
15.3.5.2.2 Potential Benefits of GoF to pandemic risk assessments: utility and limitations of using
molecular marker data
GoF approaches have potential to improve the accuracy, timeliness, and quantity of phenotypic
information generated by inspecting sequences for the presence of molecular markers for mammalian
adaptation, transmissibility, and virulence. This section focuses on the utility and limitations of molecular
marker data to the pandemic risk assessment process, relative to other types of data that are considered
(e.g. virological data, epidemiological data, and ecological data).
1446 Homeland Security Council. (2005) National Strategy for Influenza. Washington, D.C.

Molecular marker data
The genomic variation risk element includes consideration of the genetic diversity of animal influenza
viruses, which includes the presence of known molecular markers for phenotypic properties of
concern.1449 Markers for phenotypes underlying mammalian adaptation, transmissibility, and virulence are
considered most heavily,1450 in conjunction with structural modeling to account for differences in genetic
context, if appropriate. As described above, these analyses complement results from laboratory-based
phenotypic assays, particularly in cases when clinical isolates can be directly sequenced. The major
strength of this analysis is that sequence data are now the fastest, most reliable data produced at NICs and
other field laboratories where animal influenza viruses of concern are circulating. For example, the
Chinese government uploaded the sequences of the viral isolates from the first three human cases of
H7N9 influenza promptly, before additional information about the phenotypic properties of the virus was
available. The US CDC received the wild type virus from China 12 days later, after which additional
phenotypic testing could begin, resulting in a lag time for production of phenotypic data of several weeks
relative to genetic data.1451,1452 However, the predictive value of molecular markers is compromised by
significant sources of scientific uncertainty associated with the functional generalizability of the markers
and the linkage between underlying phenotypes and adaptation/transmissibility/virulence, as described
above. Because of these uncertainties, molecular marker data contributes moderately to the risk
assessment, relative to other factors. For example, in the three-virus relative risk assessment referenced
above, findings related to epidemiology risk elements were about six-fold more important than findings in
the genomic variation risk element. GoF approaches have the potential to improve the predictive value of
molecular markers, but whether that will translate to an increased weight relative to other factors
considered in the risk assessment is unknown.
15.3.5.2.3 Potential benefits and limitations of alternative pandemic risk assessment factors
Virologic data
The relative strengths and weaknesses of using molecular markers versus virological approaches to
characterize the phenotypic properties of surveillance viruses were discussed extensively in section
15.3.4. This section evaluates the utility and limitations of virologic data in the context of the overall
pandemic risk assessment.
Several risk elements rely on laboratory data: receptor binding (preference for “human-like” α2,6
sialylated receptors, “avian-like” α2,3 sialylated receptors, or dual specificity), transmission in animal
models, antiviral resistance, disease severity in animal models, and antigenic relationship between virus
and existing CVVs/vaccines.1453,1454 Although epidemiologic measurements also provide information
about the severity and transmissibility of a virus, these phenotypes are difficult to measure accurately in
nature, especially when a virus first emerges in human populations and epidemiological data are scarce.
As performing human transmission and virulence studies using novel influenza viruses would be
unethical, laboratory-generated phenotypic data critically complement epidemiologic observations.
Accordingly, in a recent assessment of three influenza viruses (an avian H1N1 virus, a human isolate of
H7N9, and a human isolate of H3N2v), these elements were highly weighted. For evaluating the
1451 (2015p) Interview with CDC Representative.
1452 Dormitzer PR. (2014) Synthetic Influenza Vaccine Viruses. Session 5. National Academy of Sciences Symposium on
Potential Risks and Benefits of Gain of Function Research

likelihood of emergence, transmission data were approximately two-thirds the value of data about the
extent of human infections (the highest-value element), and receptor binding data were half the value of
the human infection data. For evaluating potential consequences of emergence, disease severity was the
most important risk element. (The disease severity risk score reflects the severity of human infections and
the severity in appropriate animal models.)1455 The major limitations associated with reliance on
laboratory-generated phenotypic data were described above. In sum, the virus composition and/or
sequence may change during the isolation process, such that assay results do not accurately reflect the
characteristics of the viral species present in the original clinical sample, and political, logistical, and
regulatory factors delay receipt of clinical specimens/viral isolates in US labs.
Epidemiologic data
Three risk elements rely on epidemiologic data: human infections, disease severity (which is also
informed by laboratory testing in animals), and population immunity (detection of pre-existing cross-
reactive serum antibodies). The human infections and disease severity elements are the most important
elements of the likelihood and consequences components of the IRAT, respectively, because the data
directly reflect the properties of the virus in humans. However, there are several challenges associated
with the interpretation of epidemiological data for pandemic risk assessments. When a novel virus first
emerges, extrapolating virus properties from a limited number of human cases may be difficult. In
particular, disease severity is often initially over-estimated because only severe cases interact with the
public health system, and serological studies to ascertain population exposure are difficult and time-
consuming to carry out.
Ecological/environmental factors
Finally, two risk elements involve ecological factors, which collectively consider the global distribution
of the virus in animals: the number of species that can be and are infected, and the potential extent of
exposure between humans and those animal species. Other environmental information, such as the
strength of the public health systems and the strength of the relationship between the public health and
veterinary services sectors in countries in which the virus is circulating in animal populations, may also
be considered. These elements are moderately important in the likelihood component and minimally
contribute to the consequence component of the IRAT. Importantly, these elements reflect completely
different aspects of risk than the elements based on phenotypic, genetic, and epidemiologic data.
15.3.5.2.4 Summary – benefits of GoF approaches to pandemic risk assessments
GoF approaches have potential to benefit pandemic risk assessments by strengthening the predictive value
of molecular markers for mammalian adaptation, transmissibility, and virulence, which are a component
of the “genomic variation” risk element considered in the assessment. The relative importance of this
element relative to other risk elements places a qualitative “upper bound” on the potential benefits of GoF
research to pandemic risk assessments. Notably, because molecular marker data are currently
incorporated into pandemic risk assessments, the benefits of GoF-derived improvements to the reliability
of molecular marker data could be immediate.
The strengths and weaknesses of different types of data considered in a pandemic risk assessment are
summarized in Table 15.19, below. Epidemiological data (alt-GoF) represent the most important input to
the risk assessment, for both the likelihood and consequences of emergence component of the IRAT.
Laboratory data about transmissibility and virulence in appropriate animal models and receptor binding

specificity also significantly contribute to the overall pandemic risk score. Genomic variation, which
includes consideration of molecular marker data for mammalian adaptation, transmissibility, and
virulence, is relatively less important. Given the caveats associated with epidemiological and virological
data, subject matter experts involved in the pandemic risk assessment process emphasized the value of
corroborating information about infectivity, transmissibility, and disease severity in humans or
appropriate animal models with molecular marker data.1456 Those genetic data can increase confidence in
an estimate of risk adds certainty to decision-making downstream of the risk assessment, which is
valuable.
Molecular marker data play a more important role in the risk assessment when a novel influenza virus
first emerges in the human population. In this scenario, epidemiological data will be scant and sequence
data are likely to be available before phenotypic data, as happened when avian influenza H7N9 emerged
in China in March 2013. As a result, the use of molecular marker data enables a rapid risk assessment of
the emerging virus, so that downstream response actions can be initiated more quickly if deemed
appropriate. For example, a rapid risk assessment of H7N9 triggered the decision to immediately develop
a candidate vaccine virus. Of note, this risk assessment was also influenced by epidemiological
observations – namely, that multiple cases were reported in a short period of time, which hints at an
outbreak and possible detection issues. This rapid assessment resulted in initiation of vaccine production
three to four weeks earlier than if decision-makers had waited until complete phenotypic data were
available. Specifically, the wild type H7N9 virus arrived at the US CDC from China 12 days after the
sequences were published online, and characterizing the transmissibility and virulence of the virus in
ferrets requires an additional one to two weeks. (Of note, experts “re-ran” H7N9 through the IRAT once
phenotypic data had been generated, and the final score was relatively close to the initial score.) In the
event of a pandemic, such a three to four week head start on vaccine production could significantly reduce
pandemic-associated morbidity and mortality. For example, researchers estimate that deployment of
vaccine two weeks earlier during the 2009 H1N1 pandemic would have prevented an additional ~600,000
cases (an approximately 60% increase in the number of cases prevented), while deployment of the
vaccine four weeks earlier would have prevented an additional 1.4 million cases (an approximately 135%
increase in the number of cases prevented).1457
International surveillance for influenza is improving, especially in the wake of the 2009 pandemic, but
gaps remain, particularly in certain regions of the world (e.g. parts of Africa, regions experiencing
political instability, etc.). The limited breadth of available surveillance data constrains the potential
benefits of using pandemic risk assessments to guide decision-making about pandemic preparedness
investments. That is, experts can evaluate and prepare for pandemics caused by strains they know about
only. Mild disease cases, cases in remote areas, or cases in regions without strong surveillance and
disease reporting systems are likely to be missed by existing passive surveillance systems for novel
influenza cases. The viruses that cause these “hidden” cases could pose risks to human populations, in
which case the public would benefit from pandemic preparedness initiatives targeting those viruses.
Additionally, an improved ability to detect mild cases caused by known high-risk viruses, such as H5N1,
would increase the accuracy of risk assessments for these viruses by strengthening the quality of the
underlying epidemiological data. For these reasons, all stakeholders interviewed for this report, including
influenza researchers, public health personnel, and USG public health policy representatives, agreed that
there is a clear need to strengthen and expand influenza surveillance networks. Importantly, expanded
surveillance alone is not sufficient to improve pandemic risk assessments without concomitant
improvements to the tools used for pandemic risk assessments, including the use of molecular marker

data. Thus, strong surveillance networks function as a co-factor that is needed for the full realization of
GoF benefits to pandemic risk assessments.
As discussed in section 15.3.4. GoF approaches can also benefit surveillance for animal influenza viruses
by enabling the development of rapid assays for phenotypes underlying mammalian adaptation,
transmissibility, and virulence, as well as by improving computational models for sequence-based
predictions of underlying phenotypes. Either type of data could be used to corroborate information about
transmissibility and virulence gleaned through ferret experiments. Given the variability inherent in animal
experiments, in part because ferrets used for testing in different locations are genetically diverse, data
about underlying phenotypes could strengthen the robustness of this phenotypic information. If the
linkage between an underlying phenotype and adaptation/transmissibility/virulence is sufficiently strong,
the underlying phenotype could be used as an individual component of the risk assessment, akin to the
current sialic acid receptor binding specificity element. Both rapid phenotypic assays and computational
models could inform evaluation of this kind of risk element. The fact that weights for the sialic acid
receptor binding specificity, transmissibility, and disease severity elements are intermediate to high
suggests that validated rapid phenotypic assays could add significant value to the pandemic risk
assessment. However, the timeline for realization of this benefit is likely to be long-term. The benefits
arising from rapid phenotypic assays depends on the discovery and validation of suitable underlying
phenotypes and the development and validation of an appropriate rapid phenotypic assay. The benefits
arising from the use of computational models depend on the development of reliable models, which will
likely prove to be a significant scientific challenge. The timescales for these scientific and technical
innovations are unknown.

Benefits to decision-making in public health policy – Inform pandemic risk assessments of circulating animal influenza viruses
• Corroborate laboratory data about mammalian adaptation,
transmissibility, and virulence
GoF #1: o Increases certainty in decision-making downstream of • Predictive value of molecular markers is
the pandemic risk assessment currently limited due to several sources of
Genomic variation risk element:
• Enables rapid risk assessment of newly emerged viruses scientific uncertainty
• Information about molecular markers o Moderate contribution to overall risk score
for mammalian adaptation, o Sequence data are typically the fastest and most
reliable data available from diagnostic laboratories (e.g. five- to six- fold less important than
transmissibility, and virulence
where animal influenza of concern are circulating epidemiology data)
• Information about reassortment
o Provides a head start on pre-pandemic vaccine
development and other pandemic preparedness
activities
• Results in animal models may not translate to

Alt-GoF #1: • Provides a direct readout of infectivity, transmissibility, human disease
Virological data: and virulence in appropriate animal models
• Logistical, political, and regulatory factors
• Information about transmissibility and o Critical complement to epidemiological observations delay sample shipment to WHOCCs and
disease severity in ferrets • High contribution to overall risk score subsequent generation of phenotypic data
• Information about sialic acid receptor o About two-thirds as important as epidemiology data, o Data may not be available until after
binding specificity the most important element sequence data
Alt-GoF #2: • Data directly reflects the properties of the virus in

humans
Epidemiology data: • Reliable measurement of epidemiological
• Highest contribution to overall risk score, out of all risk
• Information about the number and elements considered
factors when new viruses first emerge in
severity of human infections human populations is difficult
• Information about pre-existing immunity in the o Early data may be incomplete and/or
• Information about the degree of pre- population complements information about properties of
existing immunity in human inaccurate
the virus and ecological factors
populations

Benefits to decision-making in public health policy – Inform pandemic risk assessments of circulating animal influenza viruses
Alt-GoF #3: • Information about ecological factors complements
information about properties of the virus and of the host • Minimal contribution to potential
Ecological data:
population consequences of virus emergence in human
• Information about the global populations
distribution of the virus in animal • Moderate contribution to likelihood that a virus will
emerge in human populations • Gaps in surveillance in animal populations
populations and the nature of human
compromise accuracy of information
exposure to infected animals

15.3.5.2.5 Public health impacts of pandemic risk assessments
Formal pandemic risk assessments are carried out to help prioritize resources for investments in pre-
pandemic vaccine development. Informal risk assessments may also guide investments in other pandemic
preparedness initiatives, such as sending a team of CDC experts abroad to investigate a concerning cluster
of zoonotic influenza infections in humans.
Strain-specific diagnostics are not developed in response to pandemic risk assessments (formal or
informal). The process for developing influenza diagnostics is well-established, and developing new
diagnostics is rapid and requires minimal resources relative to investments in pre-pandemic vaccine
development.1458,1459 A single human infection with a novel influenza sub-type is sufficient to trigger the
CDC to design primers and probes for a new diagnostic assay, and epidemiological data (i.e. the number
and severity of infections) also govern whether the CDC will undertake validation and subsequent FDA
licensing of the new assay.1460 Pandemic risk assessments do not trigger enhanced influenza surveillance
in the US either. The US public health system already has a surveillance system in place for detection of
novel influenza A infections, which must be reported to the CDC within 24 hours.1461
GoF approaches contribute to decision-making about pandemic preparedness activities insofar as

molecular marker data informs pandemic risk assessments. Thus, the value of GoF-derived data relative
to alternative factors that contribute to the risk assessment is the same as described for pandemic risk
assessments, above. Independently of a pandemic risk assessment, GoF approaches also contribute to the
selection of viruses used as the basis of pre-pandemic vaccines. Notably, completely different strategies
may also achieve the same ultimate public health goals as pre-pandemic vaccine development and testing
antiviral efficacy against high-risk strains. Below, the contribution of GoF approaches to decision-making
related to pre-pandemic vaccine development and testing antiviral efficacy against high-risk strains is
evaluated, as well as alternative approaches that aim to achieve the same public health goals.
Pre-pandemic vaccine development
Because existing influenza vaccines are strain-specific, pre-pandemic vaccines are developed to target
particular groups of high-risk strains. Depending on the overall level of risk associated with a particular
virus, the U.S. government will fund development of a pre-pandemic vaccine through various stages of
the vaccine production pipeline. Each of the following steps requires an escalating expenditure of
resources: CVV development, conduct of pre-clinical vaccine studies in animals, manufacture of clinical
trial lots of vaccine, conduct of human clinical trials, stockpiling of vaccine, and priming the population
against the novel influenza virus (e.g. administering vaccine in advance of a pandemic).1462 Collectively,
these investments will increase the availability of vaccines during a pandemic. Developing pre-pandemic
CVVs could save up to nine weeks (the time needed to develop and test a CVV), developing a vaccine
seed strain could shave off another two to three weeks, and carrying out pre-clinical studies in animals or
1458 Diagnostic assays for animal influenza viruses are real-time PCR-based. Diagnostic targets include the M gene (a generic
marker for influenza A viruses) and the HA gene (for sub-typing), and may also include the NA gene. The development of a
new diagnostic assay simply requires designing primers and probes for these genes.
1459 2015a) Interview with U.S. CDC (Centers for Disease Control) representative.
1460 Ibid.
1461 Council of State and Territorial Epidemiologists. CSTE List of Nationally Notifiable Conditions.
https://c.ymcdn.com/sites/cste.site-ym.com/resource/resmgr/CSTENotifiableConditionListA.pdf. Last Update August 2013.

human clinical trials could shorten production timelines by as much as 12 to 14 weeks.1463 Farther down
the vaccine production pipeline, stockpiling bulk antigen would allow for near-immediate deployment of
vaccine following emergence of a pandemic strain, while priming the population provides advanced
protection. In addition, manufacturers’ experience with production of the vaccine would likely streamline
subsequent large-scale production during the pandemic.1464 Although the pre-pandemic vaccine strain is
unlikely to exactly match the strain that emerges to cause a pandemic, use of adjuvants and prime-boost
regimens broaden the protection that can be achieved using a strain-specific vaccine, such that pre-
pandemic vaccines are highly likely to provide some level of protection against infection with a similar
strain.1465,1466,1467,1468,1469 Notably, resources limit the scope of the USG’s investment in pre-pandemic
vaccines, highlighting the need for strategies to prioritize vaccine development for the many influenza
viruses circulating in nature that have spilled over into human populations.1470
As described above, molecular marker data (derived from GoF approaches) may play an important role in
the decision to develop a CVV for an animal influenza virus, though decisions about downstream stages
of the vaccine production pipeline such as production of clinical lot material are likely to be delayed until
virological data are available for consideration in the risk assessment.1471 Once the decision is made to
develop a CVV, multiple strains may be available to serve as the basis for the CVV. In the event that
these strains have similar epidemiological and virological characteristics, the presence and type of
molecular markers for mammalian adaptation, transmissibility, and virulence can serve to differentiate
between strains. For example, the presence of markers associated with airborne transmissibility between
ferrets supported the decision to develop a CVV for a particular H5N1 strain among several options, in
response to an abrupt rise in the number of human cases in Cambodia in 2013.1472,1473,1474 Thus, the
application of molecular marker data enabled more granular decision-making than would have been
possible based on other data sources alone, which is valuable because resource limitations constrain that
number of CVVs that can be produced. This constraint is due to the fact that the number of facilities that
can produce pre-pandemic CVVs using Good Manufacturing Processes (GMP) is limited and that CVVs
used for vaccine production must undergo extensive safety and characterization testing, which is
resource-intensive.1475
1463 (2015r) Rapid Medical Countermeasure Response to Infectious Diseases: Enabling Sustainable Capabilities Through
Ongoing Public- and Private-Sector Partnerships: Workshop Summary. The National Academies Press.
1465 Ibid.
1466 (2015s) Influenza Vaccines. Interviews with Public Health Professionals Involved in Preventing and Responding to
1472 Schultz-Cherry S et al (2014) Influenza Gain-of-Function Experiments: Their Role in Vaccine Virus Recommendation and
Pandemic Preparedness. MBio 5
1474 Rith S et al (2014) Identification of molecular markers associated with alteration of receptor-binding specificity in a novel
genotype of highly pathogenic avian influenza A(H5N1) viruses detected in Cambodia in 2013. Journal of virology 88:
13897-13909

Notably, surveillance efforts for animal influenza viruses include cases of human disease and, to a lesser
extent, surveillance of agricultural animal and wildlife populations. Both animal influenza viruses isolated
from human infections as well as animal influenza viruses that have not yet caused human infections can
be subjected to a risk assessment (formally or informally). However, because of the expense involved in
each step of pre-pandemic vaccine production, none of the above steps are likely to be undertaken unless
multiple human infections have occurred.1476 As a result, although GoF approaches may aid the
interpretation of surveillance data from animals, this proximal benefit will not lead to downstream
investments in pre-pandemic vaccine development but rather is limited to deepening understanding of the
risk associated with particular viruses. Utilizing animals as sentinels for human infections will require
substantial expansion of animal influenza surveillance networks, as well as an increased understanding of
how influenza viruses evolve in agricultural animal populations (in particular, the role of animal
vaccination) and factors that govern evolutionary dynamics at the animal-human interface.
Several completely different strategies can increase the availability of vaccines during a pandemic, thus
achieving the same ultimate public health goal. These strategies are described in detail in section
15.2.4.3.3 and are briefly summarized here. First, a universal or broad-spectrum flu vaccine could be
deployed in advance of a pandemic or could be rapidly deployed following the emergence of a novel
pandemic strain. However, influenza and vaccinology experts disagree about the scientific feasibility of
developing a universal vaccine, and one expert felt that a ten to twenty year time frame for development
is optimistic. Second, several scientific and technical advancements could shorten production timelines
for strain-specific vaccines, which would lead to faster vaccine availability during a pandemic. New
vaccine platforms, such as recombinant vaccines, can be rapidly scaled up and have shorter production
timelines than egg- and cell-based vaccines. However, the one recombinant vaccine on the market
accounts for less than 1% of total seasonal influenza vaccine produced annually, and although several
other virus-free vaccine platforms are in development, the length and expense of licensure processes for
new vaccines will delay their widespread availability. Incorporating adjuvants into existing egg- and cell-
production of the same number of doses in a shorter period of time. However, only one US-licensed
pandemic vaccine includes adjuvants. Although an active area of research, adjuvanted vaccines must
undergo standard FDA licensing procedures for new vaccines and thus are unlikely to be broadly
available in the near future. Finally, GoF research that enhances virus production enables the development
of higher-yield CVVs, which shortens vaccine production timelines by increasing the rate of bulk antigen
production. Although this research can be immediately applied to improve vaccine production, this
strategy provides the greatest benefit to the production of vaccines using poor-growing CVVs. However,
as any strain may unexpectedly generate a low-yield CVV, such as the 2009 H1N1 pandemic strain, this
benefit could significantly alleviate morbidity and mortality in the event that future pandemic strains are
also grow poorly.
Field investigations of clusters of zoonotic influenza infections abroad
The CDC participates in missions to investigate zoonotic influenza cases or clusters of concern abroad, in
conjunction with the WHO, OIE, Food and Agricultural Organization of the United Nations (FAO), and
local Ministries of Health. The goal of these missions is to supplement foundational surveillance with in-
depth investigations of ecological and environmental factors that may be contributing to spillover,
including sources of human exposure to animal influenza viruses, whether and to what extent the virus is
circulating in local animal populations, retrospective investigations of poultry deaths, and other factors.
Collectively, these data improve understanding of the risk posed by the zoonotic influenza virus in that
environment, which informs decision-making about other prevention and preparedness activities (such as
1476 (2015c) Interview with USG representative involved in pandemic risk assessment and decision-making about investments
pandemic preparedness initiatives.

whether to develop a pre-pandemic CVV). Recent examples include missions to Cambodia to investigate
an abrupt rise in human H5N1 infections in 2013, to China in 2013 to investigate the initial wave of
H7N9 human infections, and to Cairo, Egypt in March of 2015 to investigate the dramatic increase in the
number of human cases of H5N1 infection recorded at the end of 2014 leading into the first few months
of 2015.1477,1478 The decision to send a CDC team abroad is informed by an assessment of whether the
sequences of human isolates contain molecular markers for mammalian adaptation, virulence, and
transmissibility. Similar to a formal risk assessment, this decision is driven by epidemiologic data but the
presence of molecular markers of concern increases adds value by increasing certainty in decision-
making. In addition, consideration of molecular marker data may stimulate increased attention to
investigations of the local animal population and human interactions with infected animals, undertaken to
better understand how ecological and environmental factors are influencing the evolution of the virus in
that area.
15.4 Influenza Viruses: Detailed Analysis of the Benefits of GoF Research that Enhances
Virulence
enhance the morbidity or mortality of influenza viruses in appropriate animal models. In this section, we
provide an overview of GoF approaches in this phenotypic category and describe the scientific outcomes
and/or products of each approach.
15.4.1.1 Serial passaging of viruses in cell culture or animal models
Serial passaging of viruses in cell culture or animals selects for viruses with enhanced fitness or virulence,
respectively. This approach is performed for two purposes: (1) to develop animal models for studying the
mechanistic basis of flu-associated morbidity/mortality and for medical countermeasure development, and
(2) to identify mutations that are associated with enhanced fitness/virulence, which provides a foundation
for follow-up studies that investigate the mechanistic basis of pathogenicity. These studies can also
provide insight into host mechanisms underlying disease pathology by correlating host immune responses
with morbidity and mortality measures. Both types of serial passaging studies are performed with
seasonal and animal (i.e. avian and swine) viruses, and animals such as mice, ferrets, and swine may be
used.
15.4.1.2 Forward genetic screen to identify mutations that enhance fitness/virulence
fitness/virulence or comprehensive reassortment of parental gene segments from two viruses, followed by
characterization of the fitness or virulence of mutants in appropriate mammalian model systems to select
for mutant viruses with enhanced fitness/virulence. Sequencing emergent viruses enables the
identification of mutations or gene segments that enhance the fitness/virulence of viruses, which provides
a foundation for follow-up studies that investigate the mechanistic basis of pathogenicity in mammals.
These studies are performed using human seasonal viruses, the 1918 H1N1 pandemic virus, and animal
viruses. A variant of this approach involves the use of strains with impaired fitness due to the evolution of
antiviral resistance, to determine whether strains can recover fitness through the acquisition of
compensatory mutations, which has been performed using seasonal strains.

15.4.1.3 Targeted modification of viruses to introduce traits that are expected to enhance
fitness/virulence in mammals

introduce genetic traits that are expected to enhance the fitness/virulence of viruses followed by
characterization of the fitness/virulence of mutants in cell culture or animal model systems, respectively,
may lead to the generation of viruses with enhanced fitness/virulence in mammals. This approach is
phenotypic trait, such as evasion of a particular innate immune response, contributes to the complex
phenotype of pathogenicity, and (2) to confirm that a particular mutation or gene segment is necessary
and sufficient to enhance the fitness/virulence of viruses in appropriate model systems. Traits that are
associated with enhanced pathogenicity may be discovered through GoF approaches, such as serial
passaging, or alt-GoF approaches, such as random mutagenesis followed by screening for attenuated
virulence (loss-of-function). As above, this information provides a foundation for follow-up studies
investigating the mechanistic basis of pathogenicity. These studies are performed using human seasonal
viruses, the 1918 H1N1 pandemic virus, and animal viruses.
15.4.2 Overview of the potential benefits of GoF experiments involving coronaviruses
Here we evaluate whether any of the GoF Influenza approaches have the potential to benefit each of the
Assessments of Gain-of-Function Research.” We also describe additional benefit areas we identified
during our research. Each potential benefit will be analyzed in detail below.
GoF approaches have the potential to benefit scientific knowledge in several ways. First, GoF approaches
provide insight into the mechanistic basis of pathogenicity, including the identification of viral and host
traits that contribute to pathogenicity. Second, information about compensatory mutations that rescue the
growth of antiviral resistant strains provides a foundation for follow-up studies investigating the
mechanistic basis of the enhanced growth phenotype, thereby benefiting scientific knowledge about the
mechanisms underlying recovery of fitness in attenuated strains as well as the mechanistic interplay
between different virus phenotypes. Finally, viruses with enhanced virulence developed using GoF
approaches can be used as tools to understand how the host immune response contributes to morbidity
and mortality observed during influenza infections, representing an indirect benefit of GoF approaches to
scientific knowledge.
GoF approaches that lead to the identification of molecular markers for enhanced pathogenicity have the
potential to inform the interpretation of wildlife, agricultural animal, and public health surveillance
information. Specifically, determining the presence (or absence) of particular mutations or of amino acid
substitutions are particular sites is one aspect of evaluating the risk posed by circulating animal influenza
viruses. Risk assessments based on evaluation of genetic surveillance data, as well as other types of data,

then inform decision-making related to public health preparedness for novel influenza outbreaks, as
discussed below.
GoF approaches that lead to the identification of compensatory mutations that rescue the fitness of
antiviral-resistant strains with impaired growth do not benefit surveillance. Because of the high mutation
rate of influenza viruses, influenza surveillance experts expect that antiviral resistant strains that initially
exhibit impaired fitness can readily acquire compensatory mutations that rescue growth. Thus, experts
simply track the presence of antiviral resistance markers, and the additional presence of absence of a
known compensatory mutation does not increase or decrease the level of risk associated with the antiviral
resistance marker.
15.4.2.3 Vaccines
GoF approaches have potential to benefit the development of vaccines in three ways:
• Serial passaging of candidate live attenuated vaccine strains in animals is used to test whether
strains recover virulence upon growth in vivo, which is an important aspect of vaccine safety.
• GoF approaches enable the identification of conserved virulence determinants in the HA and NA
proteins. These markers may be removed vaccine viruses through targeted deletion or
mutagenesis, as is commonly done for the multi-basic cleavage site present in the HA proteins
from some avian influenza strains, which may improve the efficacy and safety of the vaccine
production process.
• Viruses with enhanced virulence, generated through GoF approaches, can be used as challenge
viruses for vaccine efficacy studies, to facilitate the development of vaccines that can protect
against severe disease.
GoF approaches have potential to benefit the development of influenza therapeutics in two ways:
• GoF approaches that provide insight into viral and host traits that contribute to virulence identify
potential targets for next-generation therapeutics (either targeting the virus or the host), and
• Viruses with enhanced virulence, generated through GoF approaches, can be used as challenge
viruses for therapeutic efficacy studies, to facilitate the development of therapeutics that can
ameliorate severe disease.

GoF approaches that lead to the identification of molecular markers for enhanced pathogenicity contribute
to assessments of the pandemic risk posed by circulating animal influenza viruses, which are based on
outbreaks, including whether to develop and publicize messaging about risk factors for contracting animal
influenza infections and practices for mitigation risks, whether to enhance surveillance of animals, and
whether to develop pre-pandemic vaccines.
GoF benefits to the development of new vaccines and therapeutics could have downstream economic
benefits. We did not explicitly evaluate economic benefits in this report.
15.4.3.1 Scientific Knowledge Gap 1: What are the viral genetic and phenotypic traits that underlie
pathogenicity in mammals? What are the host factors that contribute to enhanced
pathogenicity as well as infection-associated morbidity and mortality?
15.4.3.1.1 Introduction
The pathogenesis of influenza viruses reflects the complex interactions between viral and host factors and
is the result of both the virus’s ability to cause disease and the host’s response to viral infection. From the
virus perspective, pathogenicity is a complex phenotype defined by the combined effects of many
underlying viral phenotypes including cell and tissue tropism, cytotoxicity, and replicative fitness. From
the host perspective, the immune response is essential for inhibiting viral replication, as attenuated
immune responses, such as those in immunocompromised hosts, fail to control infection. However, overly
robust responses can result in severe immunopathology. The interplay between virus and host starts with
the initiation of the early antiviral immune responses, leading to the recruitment of immune cells and the
stimulation of adaptive immunity. Unsurprisingly, influenza viruses have several mechanisms to
overcome this barrier, which contribute to fitness and pathogenicity and define the underlying phenotype
of immune evasion. Of note, NS1 performs an array of tasks that inhibit detection by the host immune
system and initiation of early immune responses, thereby providing opportunity for viral replication.
Other viral proteins that contribute to pathogenicity by immune evasion and immune antagonism include
PB1-F2, which induces host cell death and alters inflammatory responses. While advances in research
have revealed functions of specific influenza proteins and genetic traits that contribute to virulence, the
fact that overlapping and distinct mechanisms drive virulence in different strains and that a given genetic
trait or protein may exhibit distinct functions in different genetic contexts complicate the translation of
findings to other virus backgrounds. In particular, these differences pose challenges for comparing high
and low pathogenicity strains. Accordingly, much remains to be elucidated on the interplay between
virus-host interactions in defining pathogenesis.
In addition to immune evasion and antagonism, influenza viruses utilize a variety of other factors that
result in enhanced virulence. The HA protein contributes to disease severity by initiating viral attachment
and infection, and thus plays a large role in defining whether infections remain localized or become

systemic, which greatly impacts pathogenicity and disease outcomes.1480,1481,1482 Multiple aspects of HA
function contribute to virulence. For example, the HA multibasic cleavage site, found in HPAI strains,
influences tissue tropism by defining the sensitivity to tissue specific proteases that are required for its
activation during infection. Other phenotypic traits that contribute to enhanced pathogenicity include
polymerase activity and replicative fitness in mammalian cells due to adaptive mutations in the
ribonucleoprotein complex (e.g. the PB2 E627K mutation) which enables replication at the lower
temperatures observed in the human respiratory tract relative to the avian digestive tract. These examples
emphasize the complexity of the relationship between viral fitness and pathogenicity as well as the fact
that multiple, partially redundant mechanisms contribute to phenotypes underlying pathogenicity.
Considerable gaps in knowledge remain about the molecular basis and role of each underlying phenotype
in defining pathogenicity and associated disease outcomes, including systemic infection and severe
immunopathology. In particular, the relationship between fitness and pathogenicity is poorly understood,
as enhanced viral fitness in vivo does not necessarily translate to high pathogenicity. Further complicating
this field of study is the fact that the viral genetic and phenotypic traits that contribute to enhanced
virulence are not conserved in all high pathogenicity strains, suggesting that viruses may have distinct
mechanisms of pathogenesis.
The cumulative effects of the interplay between virus and host shape the pathogenicity and severity of
disease accompanying infection. Due to the complexity of the immune response and the diversity of
immune responses observed in humans, attributable to variability in underlying genetic traits, previous
exposures to influenza, and other environmental factors, there are considerable gaps in understanding how
host factors ameliorate or potentiate morbidity and mortality associated with influenza virus infection.
This is further complicated by a lack of knowledge about how early and late immune responses to
primary infection shape tissue remodeling during viral clearance and resolution of the immune
response.1483 Another knowledge gap in this area is a lack of understanding about the mechanisms
underlying patient susceptibility to and the outcomes of secondary bacterial infections, which
significantly contribute to influenza-associated morbidity and mortality. In all cases, the identification and
characterization of host factors that are necessary for viral and bacterial clearance independent of
observed immunopathology is highly sought. By differentiating between these factors, uncoupling
deleterious and protective effects of the immune response through host-targeted therapeutics may be
possible.1484
The underlying genetic and phenotypic traits that enable efficient infection and drive pathogenicity are
poorly understood, particularly because of the complex interplay among virus gene segments and between
virus and host factors. Many host and viral factors synergize to exacerbate pathology, making
mechanisms difficult to tease apart. Considerable gaps in knowledge remain about the molecular basis
and the role of each underlying phenotype in the context of the host response and viral fitness. Moreover,
there is limited understanding of the host factors that contribute to protective versus deleterious outcomes.
Insight into virus-host interactions is needed to advance in-depth understanding of virulence and
pathogenesis of influenza viruses.
1480 Bottcher-Friebertshauser E et al (2014) The hemagglutinin: a determinant of pathogenicity. Current topics in microbiology
and immunology 385: 3-34
1481 Kuiken T et al (2012) Pathogenesis of influenza virus infections: the good, the bad and the ugly. Current opinion in virology
2: 276-286
1482 Kash JC, Taubenberger JK (2015) The role of viral, host, and secondary bacterial factors in influenza pathogenesis. The
American journal of pathology 185: 1528-1536
1483 Damjanovic D et al (2012) Immunopathology in influenza virus infection: uncoupling the friend from foe. Clinical
immunology (Orlando, Fla) 144: 57-69
1484 Ibid.

Several GoF approaches can be used to discover the genetic and phenotypic markers underlying enhanced
pathogenicity of influenza viruses:
to pathogenicity by either site-directed mutagenesis or targeted reassortment (often between
animal-origin or human pandemic and human seasonal strains),

by selection for enhanced virulence, or underlying phenotypes, and
• Serial passaging in appropriate animal models or mammalian cells to select for viruses with
enhanced pathogenicity.
enhanced pathogenicity in representative model systems. The GoF approaches described here also
provide insight into host response pathways that contribute to underlying disease pathology. These
approaches can be carried out in cell culture or in animal model systems, but the former is limited to the
investigation of phenotypes underlying pathogenicity, such as replicative fitness and cell-specific immune
evasion pathways. Furthermore, these results may not translate to the complex environment and
interactions that occur during infection in vivo. The use of animal models also permits comparisons of
isolates from primary and disseminated sites of infection, as well as isolates that are shed at different time
points during infection, which can provide further insight into the genetic traits that are associated with
enhanced pathogenicity. Serial passaging has the potential to uncover novel genetic and phenotypic
markers that contribute to enhanced virulence. In contrast, because forward genetic screens involving
random mutagenesis typically focus on regions that are suspected or known to play a role in phenotypes
underlying pathogenicity, this approach can discover novel genetic markers for enhanced virulence only.
The targeted genetic modification approach is limited to the investigation of genetic traits and underlying
phenotypes that are suspected to contribute to pathogenicity (e.g. determining whether enhanced polymerase
activity contributes to pathogenicity).
necessary and sufficient to enhance virulence in mammals. Often this experiment is followed by
characterization of other virus phenotypes, such as infectivity and tissue tropism. Furthermore, this
approach provides associative insight into how host responses are altered during infection with the
modified strain. Collectively, this information provides a strong foundation for follow-up studies
investigating the mechanistic basis of pathogenicity, including the study of host-virus interactions.
Taken together, these GoF studies provide a foundation for follow-up cell biological, immununological,
and pathological studies that elucidate the mechanistic basis of viral factors contributing to virulence,
corresponding host responses, and how both factors alter susceptibility to secondary bacterial infection.
Additionally, this approach permits the identification of host immune responses that are associated with
enhanced pathogenicity. Although the analysis of host factors contributing to enhanced pathogenicity is
indirect, the information can be derived from the comparison of genetically similar virus backgrounds
displaying a dynamic range of virulence (i.e. GoF and parental strains). The relevance of these approaches
depends on whether mechanisms underlying enhanced virulence in cell culture and animal models are
representative of those in humans. This limitation may be particularly relevant for the interpretation of
studies involving mice, which are commonly used for pathogenicity studies but display distinct
pathogenesis and natural susceptibility to human influenza viruses. Alternatively, ferrets have similar
susceptibility, tissue tropism, and clinical signs of disease in response to infection with influenza viruses

as humans. Another drawback of these approaches is that results gleaned from the study of one or a few
strains may not be recapitulated in different genetic contexts.
Several alt-GoF approaches can be used to uncover genetic and phenotypic traits underlying
pathogenicity in mammals. First, comparing the sequences of human isolates that display varying degrees
of pathogenicity enables the identification of genetic changes that are associated with increased virulence.
Unlike the GoF approaches described above, this approach has the potential to directly identify genetic
traits that contribute to pathogenicity in humans and may be more likely to uncover conserved traits
through analysis of a large number of strains. However, this approach is subject to significant limitations
relative to GoF approaches. First, the success of this approach depends on the availability of a wide
breadth of surveillance data accompanied by epidemiological data about the clinical severity and case
fatality rates of particular strains or groups of strains. The fact there is considerable variability in the type
and magnitude of immune responses within human populations due to inherent genetic diversity, as well
as differences in previous exposure to influenza and vaccination status, complicates the interpretation of
genetic surveillance data. Because disease pathology can be exacerbated by host and viral factors, high-
quality “metadata” about relevant host features (e.g. age, vaccination status, etc.) is needed so that
sequences can be appropriately “binned” into low- and high-virulence categories for comparison. This is
important for both the identification of viral factors (e.g. the neurotropism observed during H5N1
infections) that may contribute to virulence as well as the identification of host factors associated with
enhanced pathogenicity (e.g. the immunopathology observed during H5N1 infections). Often, such
metadata is not provided, is incomplete, and/or is not available as quickly as genetic data in standard
surveillance practices, resulting in this approach being unfeasible or delayed relative to GoF approaches.
Second, the use of consensus sequences in standard surveillance practices may not be able to uncover
genetic traits that are present at low frequencies in human populations. Finally, the extensive genetic
diversity within circulating virus populations makes discerning distinct viral genetic traits that are likely
to contribute to pathogenicity difficult. Namely, the “noise” associated with comparing the sequences of
isolates from different patients obscures the discovery of relevant features that distinguish isolates of
varying pathogenicity, which practically limits this approach to the investigation of traits or regions
previously known to be important for pathogenicity. A variant of the surveillance-based approach
involves corroboration of sequence data with immunopathological observations from autopsies, which
provides an opportunity to identify host factors or genetic polymorphisms that are broadly associated with
severe disease.1485 In addition to the limitations described above, this approach is limited by the
availability of autopsy data and is subject to the caveat that autopsies represent late stage, lethal disease,
which may not be representative. Comparing the sequences of isolates within patients, over the course of
infection and/or from different tissue sources, represents another surveillance-based approach for
identifying genetic traits that contribute to pathogenicity. Specifically, comparing early and late isolates
during prolonged disease and comparing isolates from the primary site of infection (i.e. the upper
respiratory tract) and those from disseminated sites (i.e. lower respiratory tract), which are associated with
increased virulence, enables the identification of adaptive mutations that enhance virulence. A strength of
this approach is that the reduced viral genetic diversity observed within a single patient may enable the
identification of novel genetic traits associated with virulence. However, such traits may not be relevant
in a broader patient context due to existing diversity in human susceptibility. Moreover, this is limited to
the analysis of viral isolates from patients presenting with severe disease, which may bias findings
towards traits associated with prolonged and late stage disease.
Phenotypic characterization of wild type viruses in appropriate cell culture or animal models is another
alt-GoF approach that can be used to study mechanisms underlying pathogenicity in mammals.
1485 Everitt AR et al (2012) IFITM3 restricts the morbidity and mortality associated with influenza. Nature 484: 519-523

Specifically, comparing the sequences of wild type viruses with varied levels of fitness in vitro and
pathogenicity in vivo enables the identification of genetic and phenotypic traits associated with increased
virulence in representative cell culture or animal models, respectively. Similar to GoF approaches, this
approach can also identify host response pathways that are associated with varying disease outcomes,
including susceptibility to secondary infection. Notably, the information generated through use of cell
culture systems is limited relative to that generated through animal experiments due to the simplicity of
the host immune response in vitro. Because of the high genetic diversity among existing viral isolates
phenotypic characterization is often limited to the analysis of known determinants of pathogenicity unless
highly genetically similar strains are available. The use of in vivo models is restricted to the study of
viruses that can productively infect representative animal model systems, which excludes some animal-
origin viruses with low fitness. (Such strains are typically passaged in mice for adaptation prior to
analysis of virulence, which represents a GoF approach.) As for the GoF approaches, genetic and
phenotypic traits uncovered through this approach may not translate to humans.
modification to identify changes that attenuate fitness/virulence, can also provide information about
genetic and phenotypic traits that contribute to pathogenicity. The screening approach has the potential to
identify novel genetic traits associated with pathogenicity, while the targeted approach is used to confirm
whether particular genetic traits are necessary for pathogenicity. This information complements that
generated by GoF methods, but LoF approaches suffer from several limitations. First, because of the high
mutation rate of influenza viruses, LoF mutations that attenuate pathogenicity may revert during the
single round of passage that is needed to characterize the virulence of the mutants (which represents a
selection step). Second, although in principle, LoF screens for mutations that attenuate virulence can be
performed in an unbiased manner, characterizing the pathogenicity of a large panel of mutants in animals
is labor-intensive and expensive. As a result, the use of this method may be practically limited to cell
culture systems or the investigation viral phenotypes previously shown to be associated with
pathogenicity. Third, because many mutations attenuate pathogenicity for trivial reasons, for example
mutations that compromise viability, discovering traits that directly contribute to virulence in high
pathogenicity strains relative to low pathogenicity strains may be difficult using a LoF approach.
However, mechanistic insight into the role of non-essential virus proteins, such as PB1-F2, is feasible
using this approach, and the roles of essential proteins such as NS1 can be studied through specific
deletion of non-essential functional domains. Of note, the virulence of highly attenuated strains can still
be assessed in immunocompromised mice that are susceptible to infection, in order to identify secondary
functions that contribute to virulence, but with decreased mechanistic insight into pathogenicity.
The use of replication incompetent viruses provides another alternative method for the identification of
genetic and phenotypic traits underlying pathogenicity.1486 In these model systems, viral replication and
immune evasion pathways, both of which contribute to pathogenicity in vivo, can be assessed in cell
culture lines that are engineered to stably express an essential viral protein that is missing from the
“replication-incompetent” virus strains used for infection. For example, the replacement of the PB2 gene
with a GFP-expression construct that has the necessary flanking, non-coding and packaging sequences
from the viral genome can only replicate in cell lines that stably express exogenous PB2.1487 The result is
a virus that is biologically constrained to replication in that cell line. Several replication incompetent
model systems have been made, although some suffer from poor maintenance of the foreign gene/gene
1486 The use of this approach has been proposed during interviews with influenza researchers as a possible method, although the
use of this approach for explicitly identifying genetic and phenotypic viral and host factors contributing to fitness and cell-
specific immune evasion is currently limited.

segment (GFP) during virus packaging.1488,1489 Using these systems, viruses can be serially passaged to
identify novel adaptive mutations (and phenotypic changes) that are associated with phenotypes
underlying pathogenicity. However, cell culture systems cannot provide information about the effect of
identified genetic traits on global host responses, virus dissemination, and associated morbidity and
mortality. Accordingly, in vitro results may not be recapitulated during in vivo infection, a limitation that
further weakens the utility of this approach. An additional concern is that, due to epistasis, existing cell
lines, which express viral proteins from a particular strain, may not be compatible with other virus strain
gene segments (i.e. a cell line expressing PB2 from a lab-adapted virus such as PR8 may not be
compatible with the other gene segments of an avian influenza virus). If so, it may be necessary to
generate new constructs or cell lines, perhaps decreasing the efficiency of this approach. Further
characterization and validation of this model system will alleviate this limitation.
Several in vitro virus-free methods can be used to investigate phenotypes underlying pathogenicity. Cell
biological assays (e.g. measuring polymerase activity or IFN-induction) and crystallographic
resolution of the structures of viral protein interactions with other viral or host factors (e.g. virus-host
protein-protein complexes) can provide insight into the mechanistic and biophysical basis of underlying
phenotypes. Comparative sequence analysis of viral proteins with different phenotypic properties can then
enable the identification of mutations that are associated with relevant phenotypic changes or provide
insight into the molecular basis for virus-host interactions. Alternatively, forward genetic screens can be
used to identify novel genetic traits that contribute to underlying phenotypes, while targeted modification
of viral gene segments in isolation confirms the set of genetic changes that are necessary and sufficient to
alter an underlying phenotype. Though the simplicity and relatively high-throughput nature of these
methods renders them appealing as a screening approach for the discovery of novel genetic traits
associated with pathogenicity, these approaches are inherently limited to the investigation of previously
identified viral phenotypes. An additional drawback is that results gleaned from studying the behavior of
a viral protein or phenotype in isolation may not be recapitulated in the context of the full virus or in vivo.
Although fairly rapid phenotypic assays have been developed for the study of phenotypic traits known to
be associated with pathogenicity, assays to study certain phenotypic traits may be unreliable or
unavailable for future phenotypes of interest.
The use of in silico approaches to model the biophysical properties of viral proteins, virus-host and virus-
virus protein complexes can be used to evaluate mutations that may alter phenotypes underlying
pathogenicity. For example, results from modeling the glycosylation patterns of seasonal and pandemic
HA proteins can be used to predict the susceptibility of different HA molecules to neutralization and
inhibition by host immune proteins (e.g. collectins).1490,1491 Although this approach may provide insight
into the biophysical basis of interactions underlying phenotypes of interest, the success of the approach is
limited by the accuracy of existing models.
Finally, because pathogenicity reflects virus-host interactions, several alt-GoF approaches focus on
identifying and characterizing host factors that are associated with pathogenicity, which may provide
indirect insight into viral mechanisms underlying virulence in representative animal models. The use of
transcriptional (e.g. qRT-PCR, microarray) and translational (e.g. ELISA) expression profiling, as well as
immunophenotyping (e.g. identifying the type and kinetics of immune cell recruitment) and
the Detection of Influenza Virus Neutralizing Antibodies. J Virol 84: 2157-2163
1489 Rimmelzwaan GF et al (2011) Use of GFP-expressing influenza viruses for the detection of influenza virus A/H5N1
neutralizing antibodies. Vaccine 29: 3424-3430
1490 Sun X et al (2013) N-linked glycosylation of the hemagglutinin protein influences virulence and antigenicity of the 1918
pandemic and seasonal H1N1 influenza A viruses. Journal of virology 87: 8756-8766
1491 Job ER et al (2010) Pandemic H1N1 influenza A viruses are resistant to the antiviral activities of innate immune proteins of
the collectin and pentraxin superfamilies. Journal of immunology (Baltimore, Md : 1950) 185: 4284-4291

histopathology, independently or in the context of the GoF and alt-GoF approaches discussed above, can
identify host response pathways that change during infection and thus may play a role in pathogenicity.
The use of genetically modified mouse lines (e.g. knockout mice) or pharmacological inhibitors to
confirm the role of a particular protein, signaling pathway, or immune cell type in pathogenicity provides
further insight into the role of host-virus interactions. The strength of these approaches is that they
provide direct information about host factors involved in pathogenicity. However, the immune response
to influenza viruses is poorly understood and quite complex, making it difficult to resolve the function of
particular host proteins in the context of globally altered host factors and regulatory networks.
Given the complexity of the immune response to influenza viruses in animal models, a more targeted
approach involves in vitro proteomic (e.g. mass spectrometry) and genomic screens (e.g. RNAi screen)
utilizing both virus-free and cell culture-based infection systems to discover host factors that interact with
virus proteins of interest or that are critical for underlying phenotypes such as viral replication and
immune evasion. These approaches provide direct insight into host factors involved in viral fitness.
However, the breadth of proteomic approaches is limited in that screens typically focus on a single viral
protein, and both genomic and proteomic screens can identify host proteins that may not be functionally
relevant or may play minor roles in the viral life cycle in vivo. Furthermore, in vitro systems do not
effectively capture the complex host environment, so the function and importance of host factors
identified and studied in cell culture may not be recapitulated in vivo. The use of virus free, in vitro
systems is further limited to the analysis of viral phenotypes in isolation and may not be conserved in the
context of the full viral life cycle.
A second type of alternative approach involves the use of attenuated viruses, as a risk mitigation strategy.
Three types of attenuated viruses are used for such studies: (1) reassortants with surface protein gene
segments from seasonal influenza viruses, to which the general population has pre-existing immunity, (2)
reassortants with lab-adapted viruses (e.g. PR8), and (3) strains which have virulence factors altered or
deleted (e.g. deletion of the multi-basic cleavage site in HPAI HA sequences). The use of reassortants
with lab-adapted strains to identify viral determinants that are necessary and sufficient to enhance
virulence in a low-pathogenicity background is possible, as many of these strains are well characterized
and provide a large dynamic range for evaluating increases in virulence. Despite those advantages, the
results gleaned through use of attenuated viruses are subject to the caveat of epistasis. That is, because
complex, multi-genic traits depend on genetic context, causative genetic and phenotypic traits that
contribute to enhanced virulence in attenuated strains may not be recapitulated in the context of other wild
type strains and interactions with other factors (not present in the attenuated strain) may contribute to
virulence. Similarly, differences in disease pathogenesis relative to wild type viruses further compromise
the relevance of results gained through the use of some attenuated strains. Several additional factors limit
the range of information that can be generated using the risk mediation approach. First, seasonal
reassortant strains can only be used to study the role of internal gene segments in pathogenicity, while
lab-adapted reassortants are limited to the study of proteins donated by the wild type strain. Other types of
attenuated strains, such as strains in which the multi-basic cleavage site has been deleted, may not be
suitable for in vivo studies. For all of these methods, the mechanism of attenuation may alter phenotypes
underlying virulence in representative animal models compromising the relevancy of information gleaned
from the use of the attenuated strain.
Tables 15.20 and 15.21 provide a summary of the benefits and limitations of GoF and alt-GoF approaches
that can address scientific knowledge gaps about the mechanisms underlying viral virulence and disease
pathogenesis in mammals. The underlying genetic and phenotypic features that result in infectivity,
pathogenicity, and associated morbidity and mortality during influenza virus infection are poorly
understood, in part because of the complex interplay between virus and host factors during pathogenesis.

Many host and viral factors synergize to exacerbate pathology, making mechanisms difficult to tease
apart. Considerable gaps in knowledge remain on the molecular basis and the role of each underlying
viral phenotype in determining virulence. Moreover, there is limited knowledge on host factors that
contribute to protective immunity and immunopathology. Insight into virus-host interactions is needed to
advance in-depth understanding of virulence and pathogenesis of influenza viruses. By differentiating
between virus and host factors contributing to pathogenesis, it may be possible to target viral factors that
drive pathogenicity and to uncouple deleterious and protective effects of the immune response through
host-targeted therapeutics. Because GoF and alt-GoF approaches have distinct benefits and limitations for
the study of viral factors versus host factors that contribute to pathogenicity, their relative value for
identifying and characterizing virus factors versus host factors is evaluated separately.
Identification and characterization of viral factors that contribute to pathogenicity
The ability of GoF versus alternative approaches to provide insight into the viral factors governing
virulence and disease pathogenesis is first summarized. Taken together, GoF approaches represent the
most efficient and effective strategies for identifying novel viral genetic traits that contribute to the
pathogenicity of any virus strain. In addition, targeted genetic modification of viruses to introduce traits
necessary and sufficient to enhance virulence across multiple virus contexts. However, results gleaned
from cell culture and animal model studies may not translate to humans. Notably, the use of attenuated
strains for these studies is hindered by the fact attenuation may alter disease pathogenesis, thus results
may not be recapitulated in the genetic context of the wild type virus. In addition, attenuated strains
cannot be used when the mechanism of attenuation alters the viral factor or underlying phenotype studied.
However, the introduction of genetic traits associated with virulence to lab-adapted strains provides a
controlled system for the dissection of the functions of individual genetic or phenotypic traits that
contribute to virulence, and the fact that lab-adapted strains are attenuated permits investigation of a large
spectrum of virulence.
Although comparative sequence analysis of surveillance data has the potential to uncover viral genetic
traits that are associated with virulence in humans, the utility of this approach is significantly
compromised by shortcomings in the quality and availability of associated metadata, which are needed to
control for variability in the human immune response and susceptibility to influenza viruses. Additionally,
this approach is practically limited to the investigation of known viral genetic traits due to the high
genetic diversity among influenza viruses. For the same reason, characterization of wild type isolates is
limited to the study of previously known traits, unless genetically similar strains are available. In contrast,
comparative analysis of isolates within patients enables the identification of novel adaptive traits that are
associated with enhanced virulence over the course of infection. However this approach is often biased to
severe and late stage infection and is further complicated by the fact that individual genetic and
environmental host factors impacting immunity, thus results may not be broadly conserved in human
populations. LoF approaches also have limited utility for broad and unbiased identification of novel
genetic and phenotypic traits due to their inefficiency, including the fact that LoF approaches may
uncover traits that indirectly contribute to pathogenicity. Notably, targeted LoF enables the identification
of genetic and phenotypic traits that are necessary for enhanced virulence, which provides valuable
information to complement and strengthen results gleaned from targeted GoF studies.
While in vitro, virus free approaches and use of replication incompetent viruses enable the identification
of novel genetic and phenotypic traits that are necessary and sufficient to alter phenotypes underlying
pathogenicity, the importance of those genetic traits in the context of the complex host environment is
difficult to extrapolate. Moreover, the in vitro, virus free and cell culture methods do not provide any
information on mechanisms underlying the morbidity and mortality associated with influenza infection.

Finally, host-focused approaches provide indirect insight into the function of virus proteins and thus are
of limited utility for understanding how viral factors contribute to pathogenicity, relative to GoF
approaches.
Identification and characterization of host factors that contribute to pathogenicity
Both GoF and alt-GoF approaches can provide insight into host factors that enhance pathogenicity,
including deleterious immune responses that contribute to the morbidity and mortality caused by
influenza infection. GoF approaches can be used to identify host factors that are associated with enhanced
virulence and morbidity and mortality. In particular, targeted genetic modification to introduce traits that
are expected to enhance virulence provides a controlled system that can be used to tease apart the
interplay between virus and host factors contributing to pathogenesis, i.e. by demonstrating how changes
to a particular virus factor alter host immune responses and enhance infection-associated-pathology. The
utility of using risk-mediation reassortants in lieu of wild type viruses is significantly limited for the study
of host factors that contribute to pathogenicity. Pathogenicity is derived from the complex interplay
between many underlying viral traits and host factors that are not fully captured in the context of different
genetic backgrounds or when pathogenicity is severely impaired (e.g. with the use deletion of the MBCS).
The main drawback of GoF approaches, with respect to the study of host factors that contribute to
pathogenicity, is that they cannot establish a causal link between a host factor and enhanced pathogenicity
and/or more severe disease pathology. Additionally, results from representative animal models may not
translate to humans.
The use of targeted knockout animals or pharmacological inhibition of the host factor during infection, an
alt-GoF approach, is uniquely capable of confirming that a host factor contributes to virulence and
pathogenicity. However, because the host response is dynamic and complex, inhibition of a host factor is
likely to have a multi-faceted effect on immune responses during infection, making the identification of
host traits that drive viral clearance and deleterious immune responses difficult to resolve. Targeted
genetic modification of viruses to introduce traits expected to attenuate virulence (LoF) can also be used
to identify host factors/responses that are associated with enhanced pathogenicity. Like its GoF
counterpart (i.e. targeted genetic modification of viruses to introduce traits expected to enhance
virulence), this approach provides a controlled system for studying interplay between virus and host
factors contributing to pathogenesis, and the resulting information complements results from GoF studies.
However, LoF approaches have limited utility for studying host proteins that interact with viral proteins
that are required for fitness/infectivity. Immunological characterization of wild type isolates exhibiting
varied levels of virulence can demonstrate an association between a particular host response and
exacerbated disease pathology. However, this approach provides little mechanistic insight into the role of
particular virus-host interactions if viral isolates display high genetic diversity. Several other alt-GoF
approaches provide correlative data about the course of disease and the immune responses that are
associated with severe outcomes observed in humans, including comparative analysis of genetic
surveillance data, analysis of patient isolates, and analysis of autopsy data. This information is highly
valuable for connecting results observed in animal model systems to nature (e.g. whether neurotropism
observed during infections of ferrets with H5N1 viruses is representative of human infections). However,
these approaches provide limited mechanistic insight and are impaired by limitations in the quality and
availability of genetic surveillance data, in particular a lack of high quality metadata.

Table 15.20. Summary of the Benefits of GoF Approaches that Enhance Pathogenicity
Scientific Knowledge Benefits— What are the genetic and phenotypic traits of pathogenicity of influenza viruses? What are the host factors associated with
pathogenicity and influenza associated morbidity/mortality?
• Identifies viral genetic and phenotypic traits that are

necessary and sufficient for enhanced pathogenicity
(i.e. provides causative data) • Narrow breadth – Results may not generalize to other virus strains
• Identifies host factors associated with deleterious or • Translatability – Results from representative animal models may not
protective outcomes translate to mechanisms underlying pathogenicity in humans
a
GoF #1a [1] : • Gain insight into viral phenotypes underlying • Bias – Limited to investigation of previously identified phenotypic or
pathogenicity genetic traits
Targeted genetic modification to
introduce genetic changes • Gain insight into host mechanisms underlying disease • Limited/poor foundational knowledge of the linkage between
expected to contribute to pathology underlying phenotypes and pathogenicity and the interplay among and
pathogenicity (in vivo) • Controlled system for the study of how virus-host between viral and host factors provide further uncertainty
interactions contribute to pathogenicity • Lack of publication of negative results – Compromises evaluation of
• Proactive - can be performed using viruses that do not whether the function of particular markers is broadly conserved
display enhanced pathogenicity in humans • Indirect insight into host factors contributing to pathogenicity
• Enables testing of markers in different strain contexts
to assess generalizability of previous findings


necessary and sufficient for viral fitness (i.e. provides • Limited to the investigation of viral fitness and cell-specific immune
causative data) evasion pathways, which are components of pathogenicity
• Identifies host factors associated with phenotypes • Narrow breadth – Results may not generalize to other virus strains
underlying fitness/immune evasion
• Translatability – Results from cell culture models may not translate to
GoF #1b [1]: • Gain insight into viral phenotypes underlying humans
fitness/immune evasion
Targeted genetic modification to • Bias – Limited to investigation of previously identified phenotypic or
introduce genetic changes • Gain insight into host mechanisms underlying cell- genetic traits
expected to contribute to fitness specific immunity
and immune evasion (in vitro) • Controlled system for the study of how virus-host underlying phenotypes and pathogenicity and the interplay among and
interactions contribute to pathogenicity between viral and host factors provide further uncertainty
• Proactive - can be performed using viruses that do not • Lack of publication of negative results – Compromises evaluation of
display enhanced pathogenicity in humans whether the function of particular markers is broadly conserved
• Enables testing of markers in different strain contexts • Indirect insight into host factors contributing to fitness/immune evasion
to assess generalizability of previous findings
• Identifies novel genetic traits that are sufficient for • Narrow breadth – Results may not generalize to other virus strains
enhanced pathogenicity • Translatability – Results from representative animal models may not
• Identifies host factors associated with deleterious or translate to mechanisms underlying pathogenicity in humans
GoF #2a [2]: protective outcomes • Bias – Limited to investigation of previously identified phenotypic
Forward genetic screen to traits
• Gain insight into viral phenotypes underlying
introduce genetic changes that
pathogenicity • Limited/poor foundational knowledge of the linkage between
may contribute to pathogenicity,
followed by testing in vivo • Gain insight into host mechanisms underlying disease underlying phenotypes and pathogenicity and the interplay among and
pathology between viral and host factors provide further uncertainty
• Proactive - can be performed using viruses that do not • Associative – Information produced is correlative, not causative
display enhanced pathogenicity in humans • Indirect insight into host factors contributing to pathogenicity

• Identifies novel genetic traits that are sufficient to • Narrow breadth – Results may not generalize to other virus strains
enhance viral fitness/immune evasion • Limited to the investigation of viral fitness, which is one component of
• Identifies host factors associated with phenotypes pathogenicity
underlying fitness/immune evasion • Translatability – Results from cell culture models may not translate to
GoF #2b [2]:
• Gain insight into viral phenotypes underlying humans
Forward genetic screen to
introduce genetic changes that fitness/immune evasion • Bias – Limited to investigation of previously identified phenotypic
may contribute to phenotypes • Gain insight into host mechanisms underlying cell- traits
underlying pathogenicity, specific immunity • Limited/poor foundational knowledge of the linkage between
followed by testing in vitro underlying phenotypes and pathogenicity and the interplay among and
display enhanced pathogenicity in humans between viral and host factors provide further uncertainty
• In vitro methods can be used to screen a larger number • Associative – Information produced is correlative, not causative
of mutants than in vivo methods • Indirect insight into host factors contributing to fitness/immune evasion
• Identifies novel genetic and phenotypic traits that are
sufficient for enhanced pathogenicity
• Identifies host factors associated with deleterious or • Narrow breadth – Results may not generalize to other virus strains
GoF #3a [3]: protective outcomes
• Translatability – Results from representative animal models may not
Serial passaging with selection • Gain insight into viral phenotypes underlying translate to mechanisms underlying pathogenicity in humans
for pathogenicity, use of animal pathogenicity
models (in vivo) • Associative – Information produced is correlative, not causative
• Gain insight into host mechanisms underlying disease
pathology • Indirect insight into host factors contributing to pathogenicity
display enhanced pathogenicity in humans

• Identifies novel genetic and phenotypic traits that are

sufficient to enhance viral fitness/immune evasion
• Identifies host factors associated with phenotypes • Limited to the investigation of viral fitness and cell-specific immune
GoF #3b [3]: underlying fitness/immune evasion
evasion pathways, which are components of pathogenicity
Serial passaging with selection • Gain insight into viral phenotypes underlying • Translatability – Results from cell culture models may not translate to
for fitness, use of cell culture fitness/immune evasion
humans
models (in vitro) • Gain insight into host mechanisms underlying cell-
specific immunity
• Proactive - can be performed using viruses that do not • Indirect insight into host factors contributing to fitness/immune evasion
display enhanced pathogenicity in humans
• Bias – High genetic diversity among influenza strains limits this

• Identifies genetic traits that are associated with approach to the examination of genetic regions already known to be
enhanced pathogenicity important
o Comparison of genetically similar viruses can result • Limited/poor foundational knowledge of the linkage between
in the identification of previously unknown genetic underlying phenotypes and pathogenicity and the interplay among and
traits that are associated with enhanced pathogenicity between viral and host factors provide further uncertainty
o Depending on the size of analysis and strength of • High diversity in host population
association some traits can be considered “causally”
Alt-GoF #1 [1]: linked o Variability in the type and magnitude of immune responses observed
in human populations due to genetic diversity, vaccination history,
Comparative sequence and o Directly translates to human disease and previous exposure to influenza
epidemiological data analysis of o Analyzes broad data sets applicable to many strains
human isolates • Limited by the quality and availability of existing surveillance data
• Identifies host factors that correlate with deleterious or o Consensus sequences may not capture low frequency mutations
protective outcomes
o High-quality metadata on relevant host factors needed to
• Gain insight into the prevalence and distribution of appropriately bin groups for comparison may not be available, is
genetic traits incomplete, or is delayed relative to sequences
o Can infer functional generalizability of genetic traits o Limited reporting of negative surveillance data
(i.e. whether they do or do not behave similarly in
different genetic contexts) • Associative – Information produced is correlative, not causative
• Reactive – Analysis of viral isolates that already exist in nature


approach to the examination of genetic regions already known to be
important
underlying phenotypes and pathogenicity and the interplay among and
between viral and host factors provide further uncertainty
• Limited availability of case studies
• Identifies genetic traits that are associated with
enhanced pathogenicity • Analysis is biased towards severe and often late stage disease
Alt-GoF #2 [2]: o Comparison of genetically related viruses can result • High diversity in host population
Comparative sequence and in the identification of previously unknown genetic o Variability in the type and magnitude of immune responses observed
clinical data analysis of human traits that are associated with enhanced pathogenicity in human populations due to genetic diversity, vaccination history,
isolates from a single patient o Directly translates to human disease and previous exposure to influenza
• Identifies host factors that correlate with deleterious or • Limited by the quality and availability of existing surveillance data
protective outcomes o Consensus sequences may not capture low frequency mutations
o High-quality metadata on relevant host factors needed to
appropriately bin groups for comparison may not be available, is
incomplete, or is delayed relative to sequences
o Limited reporting of negative surveillance data


approach to the examination of genetic regions already known to be
• Identifies genetic and phenotypic traits that are important
necessary and sufficient for enhanced pathogenicity
(i.e. provides causative data) • Limited/poor foundational knowledge of the linkage between
o Comparison of genetically similar viruses can result underlying phenotypes and pathogenicity and the interplay among and
in the identification of sufficient genetic and between viral and host factors provide further uncertainty
Alt-GoF #3 [3]: phenotypic traits • Animal influenza viruses that have poor fitness in representative animal
Characterization of wild type • Identifies host factors associated with deleterious or models provided limited insight
viruses protective outcomes • Genetic diversity of viral isolates limits the amount of in-depth
• Gain insight into viral phenotypes underlying mechanistic insight into the interplay between virus-host
pathogenicity • Translatability – Results may not translate to mechanisms underlying
• Gain insight into host mechanisms underlying disease pathogenicity in humans
pathology • Associative – Information produced is correlative, not causative


pathogenicity in humans
• Attenuated virus may recover virulence during characterization
• Identifies previously unknown genetic and phenotypic • Bias – Limited to investigation of previously identified phenotypic
traits that are necessary for pathogenicity
traits
Alt-GoF #4 [4]: • Identifies host factors associated with deleterious or
LoF forward genetic screen to
protective outcomes
underlying phenotypes and pathogenicity and the interplay among and
introduce genetic changes that
may attenuate virulence,
• Gain insight into viral phenotypes underlying between viral and host factors provide further uncertainty
pathogenicity
followed by testing in vitro or in • Triviality – May uncover mutations that indirectly attenuate
vivo • Gain insight into host mechanisms underlying disease pathogenicity, which provides limited mechanistic insight
pathology o Less of a concern if targeting specific domains or regions of the
influenza genome which are not required for viability (e.g. NS1
functional domains)
• Practically limited to the use of cell culture systems
• Ethical considerations and resources required for animal
experiments preclude screening of a large number of mutants in vivo


• Identifies genetic and phenotypic traits that are • Narrow breadth – Results may not generalize to other virus strains
necessary for pathogenicity
• Attenuated virus may recover virulence during characterization
• Identifies host factors associated with deleterious or
protective outcomes
• Bias – Limited to investigation of previously identified phenotypic or
genetic traits
Alt-GoF #5 [5]: • Gain insight into viral phenotypes underlying
pathogenicity
Targeted LoF to introduce underlying phenotypes and pathogenicity and the interplay among and
genetic changes expected to • Gain insight into host mechanisms underlying disease between viral and host factors provide further uncertainty
attenuate pathogenicity pathology
• Triviality – May uncover mutations that indirectly attenuate
• Controlled system for the study of how virus-host pathogenicity, which provides limited mechanistic insight
interactions contribute to pathogenicity o Less of a concern if targeting specific domains or regions of the
• Enables testing of markers in different strain contexts influenza genome which are not required for viability (e.g. NS1
to assess generalizability of previous findings functional domains)
• Lack of publication of negative results – Compromises evaluation of
whether the function of particular markers is broadly conserved

• Identifies genetic and phenotypic traits that are fitness/transmissibility in humans
necessary and sufficient for viral fitness (i.e. provides • Narrow breadth – Results may not generalize to other virus strains
causative data)
Alt-GoF #6 [6]: • Limited Utility – Replication incompetent systems have only been
(In vitro, replication incompetent • Gain insight into viral phenotypes underlying developed and validated for a limited number of strains
model system) Targeted genetic fitness/immune evasion o Use of existing models for other strains will depend on genetic
modification to introduce genetic • Proactive - can be performed using compatibility
changes expected to contribute to viruses/combinations of virus gene segments that do
fitness and immune evasion
• Bias – Limited to investigation of previously identified phenotypic or
not display enhanced pathogenicity in humans genetic traits
• Enables testing of markers in different strain contexts • Limited/poor foundational knowledge of the linkage between
to assess generalizability of previous findings underlying phenotypes and pathogenicity and the interplay among and


• Identifies novel genetic and phenotypic traits that are fitness/transmissibility in humans
necessary for viral fitness (i.e. provides causative data)
Alt-GoF #7 [7]: • Narrow breadth – Results may not generalize to other virus strains
• Gain insight into viral phenotypes underlying
(In vitro, replication incompetent
fitness/immune evasion • Associative – Information produced is correlative, not causative
model system) Serial passaging
with selection for fitness • Proactive - can be performed using • Limited Utility – Replication incompetent systems have only been
viruses/combinations of virus gene segments that do developed and validated for a limited number of strains
not display enhanced pathogenicity in humans o Use of existing models for other strains will depend on genetic
compatibility
• Identifies novel genetic traits that are sufficient to alter

phenotypes underlying fitness
Alt-GoF #8 [8]: • Provides insight into mechanistic basis of underlying
(In vitro, virus-free) Forward phenotypes • Simplicity of model system – Results based on the study of a viral
genetic screen to introduce • In vitro methods can be used to screen a larger number protein/phenotype in isolation may not be recapitulated in the context of
genetic changes that may alter of mutants than in vivo methods the full virus
phenotypes underlying fitness • Narrow breadth – Results may not generalize to other virus strains
• Proactive - can be performed on virus gene segments
from viruses that do not display enhanced • Bias – Limited to investigation of previously identified phenotypic traits
pathogenicity in humans
• Identifies genetic traits that are necessary and underlying phenotypes and pathogenicity and the interplay among and
sufficient to alter a phenotype underlying fitness between viral and host factors provide further uncertainty
Alt-GoF #9 [14]: • Provides insight into the mechanistic basis of • State of methodology – Relies upon phenotypic assays, which may be
(In vitro, virus-free) Targeted phenotypes underlying fitness unreliable or unavailable
genetic modification to introduce • Proactive - can be performed on virus gene segments • Lack of publication of negative results – Compromises evaluation of
genetic changes expected to alter from viruses that do not display enhanced whether the function of particular markers is broadly conserved
phenotypes underlying fitness pathogenicity in humans
• Enables testing of markers in different viral gene
segments to assess generalizability of previous findings


Alt-GoF #10 [9]: interactions contributing to fitness
(In vitro, virus-free) Structural o Provides detailed mechanistic information
studies to analyze the molecular • Proactive - can be performed using select virus gene
basis of fitness/immune evasion segments from viruses that do not display enhanced
pathogenicity in humans depending on the state of
methodology

interactions • Predictive – Does not confirm or correlate phenotypic effects in a
Alt-GoF #11 [15]: biological context
• Provides detailed mechanistic information
(In silico, virus-free) Modeling to • Simplicity of model system – Results based on the study of a viral
analyze the biophysical effects of • Proactive - can be performed on virus gene segments protein/phenotype in isolation may not be recapitulated in the context of
mutations contributing to from viruses that do not display enhanced the full virus
pathogenicity pathogenicity in humans
• Model accuracy – Utility of the approach depends on the quality of
• Enables prediction of phenotypic consequences of existing models
markers in different viral gene segments to assess
generalizability of previous findings
Alt-GoF #12 [10]c: • Identifies host proteins that may play a role in fitness • Translatability – Results may not translate to mechanisms underlying
Proteomic screen to identify host during infection fitness/transmissibility in humans
proteins that physically interact o Reveals previously unknown host factors
with viral proteins during o Reveals previously unknown host-virus interactions
infection during infection • Bias – Limited to investigation of previously identified phenotypic traits

• Provides insight into the role of particular virus-host • Limited/poor foundational knowledge of the linkage between
interactions during infection underlying phenotypes and pathogenicity and the interplay among and
• Identified host protein may not be functionally relevant or may play a
Alt-GoF #13 [11]:
minor role in viral life cycle
Genomic screen to identify host
factors that contribute to fitness • Indirect – Identification of host proteins or virus-host interactions that
contribute to adaptation/transmissibility provides indirect information
about viral genetic and phenotypic traits underlying adaptation
o Mechanistic insight may depend on prior knowledge of virus-host
interactions

• Identifies host proteins that may play a role in fitness • Limited/poor foundational knowledge of the linkage between
o Reveals previously unknown host factors underlying phenotypes and pathogenicity and the interplay among and
contributing to underlying phenotypes between viral and host factors provide further uncertainty
Alt-GoF #14 [13]:
(In vitro, virus-free) Proteomic or o Reveals previously unknown host-virus interactions • Identified host protein may not be functionally relevant or may play a
contributing to underlying phenotypes minor role in viral life cycle
genomic screen to identify host
factors that interact with • Provides direct insight into the role of particular virus- • Indirect – Identification of host proteins or virus-host interactions that
particular virus proteins and/or host interactions contribute to adaptation/transmissibility provides indirect information
contribute to fitness about viral genetic and phenotypic traits underlying adaptation
• Proactive - can be performed on virus gene segments
from viruses that do not display enhanced o Mechanistic insight may depend on prior knowledge of virus-host
pathogenicity in humans interactions
• Simplicity of model system – Results based on the study of a viral
protein/phenotype in isolation may not be recapitulated in the context of
the full virus


• Identifies and confirms host factors that contribute to
deleterious or protective outcomes • Identified host protein may not be functionally relevant or may play a
minor role in viral life cycle
Alt-GoF #15 [12]: • Gain insight into viral phenotypes underlying
pathogenicity • May be difficult to resolve function of host protein in the context of
Targeted modification of host
global alteration of the host protein
factor to alter expression or • Gain direct insight into host mechanisms underlying
function of host factors expected disease pathology • Indirect – Identification of host proteins or virus-host interactions that
to contribute to pathogenicity contribute to adaptation/transmissibility provides indirect information
• Enables testing of the role of host markers in about viral genetic and phenotypic traits underlying pathogenicity
pathogenicity in the context of infection with new viral
strains, to assess generalizability of previous findings • Mechanistic insight may depend on prior knowledge of virus-host
interactions
• Lack of publication of negative results – Compromises evaluation of
whether the function of particular markers is broadly conserved
a

Scientific Knowledge Benefits – Utility and Limitations of Using Attenuated Strains for GoF Approaches that Enhance Pathogenicity
In vivo
High Pathogenicity Strainc • Virulence studies
• Animal strain N/A In vitro
• Pathogenic reassortant • Characterization of underlying
phenotypes of pathogenicity
Genetic Context
• Complex phenotypes are multi-genic; results may not be recapitulated in the context of the wild
type virus
Limited Utility
In vivo
• Precludes study of the role of animal-origin HA and NA proteins, which are critical viral factors • Virulence studies
Risk mediation Reassortant- in pathogenicity
Seasonal influenza In vitro
Overlapping phenotypes
• Method of attenuation may alter phenotypes that contribute to virulence, thus interfering with the • Characterization of underlying
study of virulence
Altered course of disease
• Animals infected with attenuated viruses may exhibit significantly different disease pathology,
limiting the relevance of outcomes to wildtype viruses
Limited model systems
• Lab-adapted strains are not highly infectious in ferrets
Genetic Context
In vivo
• Complex phenotypes are multi-genic; results may not be recapitulated in the context of the • Virulence studies
pathogenic virus
Risk mediation Reassortant- In vitro
Lab-adapted (e.g. PR8) Overlapping phenotypes
• Method of attenuation may alter phenotypes that contribute to virulence, thus interfering with the • Characterization of underlying
phenotypes of mammalian
study of virulence
adaptation and transmissibility
Altered course of disease
• Animals infected with attenuated viruses may exhibit significantly different disease pathology,
limiting the relevance of outcomes to wildtype viruses

Scientific Knowledge Benefits – Utility and Limitations of Using Attenuated Strains for GoF Approaches that Enhance Pathogenicity
Genetic Context
In vivo
Attenuated Strain • Complex phenotypes are multi-genic; results may not be recapitulated in the context of the wild • Virulence studies
• Targeted mutagenesis to type virus
In vitro
remove virulence factor Overlapping phenotypes
(e.g. ΔMBCS) • Method of attenuation may alter phenotypes that contribute to virulence, thus interfering with the • Characterization of underlying
study of virulence
a
Animal-origin strains include avian- and swine-origin strains that have and have not infected humans. Pandemic strains include the 1918 H1N1, 1957 H2N2, and 1968
H3N2 viruses. Seasonal strains include all seasonal isolates and 2009 H1N1 pandemic isolates (now circulating seasonally). Risk mediation reassortants include all
reassortants with lab-adapted viruses or with surface protein gene segments from seasonal influenza viruses. Pathogenic reassortants include viruses with animal and/or
human gene segments (both seasonal and pandemic) for which human populations have limited or no immunity.
b
The text color in the experimental system column indicates the general feasibility of the use of the virus described for in vivo or in vitro use. Green indicates that the virus
would likely be functional and representative of wild-type conditions in vitro and in vivo, orange indicates that the virus may not be functional or representative in vitro or
in vivo, and red indicates that the virus would likely be non-functional in vitro or in vivo.
c
GoF approaches are shaded in blue, and alt-GoF approaches (i.e. conducting GoF approaches using attenuated strains in lieu of wild type strains) are shaded in grey.

15.4.3.2 Scientific knowledge gap 2: Discover whether fitness defects associated with the acquisition of
Though influenza viruses can readily mutate to acquire resistance to therapeutics, antiviral-resistant
viruses are often initially less fit than parental viruses.1492 For example, when the H274Y mutation in the
NA gene (N1 numbering), which confers strong resistance to oseltamivir, first arose in nature, viruses
carrying that mutation were less fit than their wild type counterparts.1493 The relative fitness of antiviral-
resistant strains has implications for how likely and how quickly these strains are to spread in nature.
Whether and how antiviral strains can acquire compensatory mutations that enhance fitness while
preserving the antiviral resistance phenotype is unknown for most antiviral resistance mutations. Studies
investigating this question provide insight into the mechanistic basis of viral fitness and the mechanistic
interplay between antiviral resistance and other virus phenotypes, and also are of interest for public
health.
Several GoF approaches can be used to determine whether antiviral-resistance strains with impaired
growth can recover fitness and to identify compensatory mutations that rescue growth, which provides a
foundation for follow-up biochemical and cell biological studies that investigate the mechanistic basis of
enhanced growth. First, growth-impaired strains can be serially passaged in cells or animals to select for
strains with enhanced fitness, following by sequencing of emergent viruses to identify genetic changes
that arose. However, this approach often results in reversion of antiviral-resistance mutations rather than
the evolution of compensatory mutations. A second approach involves forward genetic screens to identify
mutations that are sufficient to rescue fitness. While this approach is more likely to uncover compensatory
mutations than serial passaging, screening large libraries of mutants is relatively labor-intensive,
particularly if mutations are introduced into multiple virus proteins (as compensatory mutations may arise
in proteins that do not contain antiviral-resistance mutations). Finally, targeted mutagenesis is used to
confirm that a particular mutation or set of mutations is necessary and sufficient to rescue the fitness of a
growth-impaired strain.
15.4.3.2.2 Benefits and Limitations of Alt-GoF Approaches
Two alt-GoF approaches can be used to identify compensatory mutations that may rescue the growth of
antiviral-resistant strains with impaired fitness. First, comparative analysis of the sequences of antiviral-
resistant strains with varying levels of fitness may enable the identification of mutations that are
associated with enhanced fitness. However, due to the high genetic diversity among influenza viruses,
generating strong hypotheses about mutations that are linked to the recovery of fitness is difficult. In
addition, this approach is reactive, limited to the discovery of compensatory mutations after antiviral-
resistant strains have recovered growth in nature. A second approach involves computational modeling to
predict mutations that may rescue the fitness of growth-impaired strains. While this approach has been
used to successfully predict mutations that enhance the growth of antiviral-resistant strains carrying the
1492 Baek YH et al (2015) Profiling and characterization of influenza virus N1 strains potentially resistant to multiple
neuraminidase inhibitors. Journal of virology 89: 287-299

H274Y mutation in NA (N1 numbering),1494,1495 all predictions must be experimentally confirmed using
targeted mutagenesis, a GoF approach. Additionally, because existing computational models cannot
predict epistasis effects, the in silico approach is limited to the discovery of compensatory mutations that
arise in the same protein carrying the antiviral-resistance mutations.
The strengths and limitations of GoF and alt-GoF approaches that can be used to investigate whether and
how antiviral-resistant strains can overcome fitness defects are summarized in Table 15.22. Taken
together, GoF approaches are uniquely capable of proactively discovering compensatory mutations that
rescue the fitness of any antiviral-resistant strain with impaired growth, as well as establishing a causal
link between compensatory mutations and enhanced fitness. Computational modeling can be used to
generate hypotheses about mutations that may rescue growth, but all predictions must be experimentally
confirmed using GoF approaches. Comparative sequence analysis of antiviral-resistant strains with varied
levels of fitness has significant limitations relative to other approaches.
1494 Bloom JD, Glassman MJ (2009) Inferring Stabilizing Mutations from Protein Phylogenies: Application to Influenza
Hemagglutinin. PLoS Comput Biol 5
(New York, NY) 328: 1272-1275

Table 15.22. Summary of the Benefits of GoF Approaches that Enhance Virulence in Mammals
Scientific Knowledge Benefits – Discover whether fitness defects associated with the acquisition of antiviral resistance can be overcome, and the
mechanisms underlying recovery of fitness
• Identifies novel compensatory mutations that are • Often results in reversion of antiviral-resistance or other
GoF #1: sufficient to rescue the growth of attenuated strains attenuating mutations rather than selection for
Serial passaging of attenuated compensatory mutations
• Proactive – can be performed on any virus strain
viruses in cells or animals • Associative - Information produced is correlative, not
causative
GoF #2:
• Identifies novel compensatory mutations that are • Screening large libraries of mutant viruses is labor-
Forward genetic screen to sufficient to rescue the growth of attenuated strains intensive
identify compensatory
mutations that rescue fitness • Proactive – can be performed on any virus strain • Information produced may be correlative, not causative
• Identifies compensatory mutations that are necessary and

GoF #3: sufficient to rescue the growth of attenuated strains
Targeted mutagenesis to across multiple strain contexts • Lack of publication of negative results – Compromises
introduce compensatory evaluation of whether the function of particular markers
• Gain insight into mechanisms underlying the recovery of is broadly conserved
mutations expected to enhance
fitness of antiviral-resistant strains
the growth of attenuated strains
• Proactive – can be performed on any virus strain
• Associative - Information produced is correlative, not
causative
Alt-GoF #1:
• Identifies compensatory mutations that are associated • Challenging due to the high genetic diversity among
Comparative sequence analysis
with enhanced growth of antiviral resistant strains with influenza viruses
of antiviral-resistant strains
attenuated fitness
with varying levels of fitness • Reactive – limited to discovering compensatory
mutations after antiviral-resistant strains have recovered
growth in nature

Scientific Knowledge Benefits – Discover whether fitness defects associated with the acquisition of antiviral resistance can be overcome, and the
mechanisms underlying recovery of fitness
• Predictive – Does not confirm or correlate phenotypic

Alt-GoF #2: effects in a biological context
Computational modeling to • Predicts compensatory mutations that may rescue the • Model accuracy – utility of approach depends on the
predict compensatory growth of antiviral resistant strains with attenuated fitness quality of existing models
mutations that will rescue the
growth of antiviral-resistant • Proactive – can be performed on any virus strain • Limited to the prediction of compensatory mutations
strains with impaired fitness within antiviral target protein
• Existing models cannot account for epistasis effects

15.4.3.3 Scientific Knowledge and Vaccine/Therapeutic Development Benefit: Development of Animal
Models
observed during human infections are essential for the study of influenza-associated morbidity/mortality
and for testing the safety and efficacy of new vaccines and therapeutics. Mice, a common animal model
used for the development of influenza MCMs, are naturally resistant to infection with many influenza
viruses. GoF or alt-GoF approaches can be used to develop animal models to study the effectiveness of
MCMs against these viruses. The development of MCMs that protect against severe disease necessitates
testing the efficacy of candidate MCMs in animal models that exhibit exacerbated disease pathology. In
cases where wild type viruses cause a limited spectrum of disease, GoF or alt-GoF approaches may be
used to generate model systems that display a larger dynamic range of virulence.
15.4.3.3.1 Potential Benefits and Limitations of GoF approaches
Serial passaging of influenza viruses in laboratory animals to generate animal models is performed for
two purposes. The first purpose is the generation of viruses that are capable of efficiently infecting mice
for the study of influenza pathogenesis and medical countermeasure (MCM) development, as mice are
inherently resistant to infection with human seasonal influenza viruses and some animal influenza viruses.
Mouse-adapted influenza viruses have been used extensively for pathogenesis studies and for testing the
efficacy of candidate vaccines and therapeutics against seasonal and pandemic influenza
viruses.1496,1497,1498,1499,1500,1501,1502,1503,1504,1505,1506 The second purpose is the generation of viruses with
enhanced pathogenicity to support the development of MCMs that are capable of protecting against
severe disease.1507 For example, this approach would facilitate testing of the protective efficacy of the
stockpiled H5N1 vaccines against the H5N2 HPAI viruses that caused widespread outbreaks in domestic
poultry populations in the spring of 2014 and continue to circulate in wild birds with sporadic spread to
poultry, which is of interest to HHS. However, North American isolates of H5N2 are of low virulence in
ferrets, and therefore cannot be used to reliably evaluate vaccine effectiveness. Limited passaging of an
avian H5N2 isolate in ferrets would select for a virus with enhanced virulence in mammals, which would
provide a more relevant assessment of the ability of the vaccine to protect against H5N2 infections in
1496 Bahgat MM et al (2011) Inhibition of lung serine proteases in mice: a potentially new approach to control influenza
infection. Virology journal 8: 27
1497 Sun K et al (2011) Seasonal FluMist vaccination induces cross-reactive T cell immunity against H1N1 (2009) influenza and
secondary bacterial infections. Journal of immunology (Baltimore, Md : 1950) 186: 987-993
1498 Droebner K et al (2011) Antiviral activity of the MEK-inhibitor U0126 against pandemic H1N1v and highly pathogenic
avian influenza virus in vitro and in vivo. Antiviral research 92: 195-203
1499 Kashyap AK et al (2010) Protection from the 2009 H1N1 pandemic influenza by an antibody from combinatorial survivor-
based libraries. PLoS pathogens 6: e1000990
1500 Leneva IA et al (2000) The neuraminidase inhibitor GS4104 (oseltamivir phosphate) is efficacious against A/Hong
Kong/156/97 (H5N1) and A/Hong Kong/1074/99 (H9N2) influenza viruses. Antiviral research 48: 101-115
1501 Govorkova EA et al (2001) Comparison of efficacies of RWJ-270201, zanamivir, and oseltamivir against H5N1, H9N2, and
other avian influenza viruses. Antimicrobial agents and chemotherapy 45: 2723-2732
1502 Ekiert DC et al (2011) A highly conserved neutralizing epitope on group 2 influenza A viruses. Science (New York, NY)
333: 843-850
1503 Moseley CE et al (2010) Peroxisome proliferator-activated receptor and AMP-activated protein kinase agonists protect
against lethal influenza virus challenge in mice. Influenza and other respiratory viruses 4: 307-311
1504 Tharakaraman K et al (2015) A broadly neutralizing human monoclonal antibody is effective against H7N9. Proceedings of
the National Academy of Sciences of the United States of America 112: 10890-10895
1505 Boon AC et al (2009) Host genetic variation affects resistance to infection with a highly pathogenic H5N1 influenza A virus
in mice. Journal of virology 83: 10417-10426
1506 Srivastava B et al (2009) Host genetic background strongly influences the response to influenza a virus infections. PloS one
4: e4857
1507 (2015h) Interviews with influenza researchers.

humans (which are expected to be accompanied by mammalian adaptive mutations in the virus that
enhance its virulence in humans).1508,1509 One key strength of the GoF approach is that the use of
genetically similar viruses that display a large range and magnitude of virulence represents a controlled
system for comparing the effectiveness of vaccines and therapeutics against low and high pathogenicity
viruses, which enables triaging of similar MCM candidates based on their ability to protect against severe
disease. The main drawback associated with these approaches is that the passaging needed to adapt the
virus to representative animal models may alter its phenotypic properties in ways that affect pathogenesis
mechanisms and its susceptibility to MCMs under study, which may render findings misrepresentative.
15.4.3.3.2 Potential Benefits and Limitations of Alt-GoF Approaches
One alternative to the use of serially passaged viruses involves sensitization of the host to influenza virus
infection. This involves increasing host susceptibility to infection through the use of inbred mouse lines,
knockout/transgenic mice, or the treatment of mice or ferrets with immunosuppressants.1510,1511,1512,1513,1514
This approach can enable the study of wild type viruses that do not efficiently infect wild type mice. For
example, although BALC/c mice are resistant to infection with many influenza viruses, the inbred DBA.2
mouse line is susceptible to infection with a variety of influenza viruses and has been used to demonstrate
the efficacy of vaccines and therapeutics against seasonal, pandemic, and animal influenza viruses, as
well as to study pathogenesis mechanisms.1515,1516,1517,1518,1519 A strength of this approach is that the
generation of genetically similar hosts (or genetically identical hosts, if immunosuppressants are used)
that display a range of disease outcomes provides a controlled system for comparing pathogenesis
mechanisms and the effectiveness of MCM candidates to protect against more severe disease. The use of
genetic modification is largely limited to the use of the mouse model system, for which there are a broad
array of well-established tools. However, the mouse model is less representative of human disease than
other animal models, such as the ferret. The use of immunosuppressants is a promising alternative. The
key drawback of this approach is that results gleaned from the use of immunocompromised hosts may not
translate to disease in healthy hosts.
A second alternative approach involves infection of wild type hosts with wild type viruses. As mice are
naturally resistant to infection with many influenza viruses, the utility of this approach is limited for the
1508 (2015o) Interview with U.S. government representative involved in influenza vaccine development.
1509 Pulit-Penaloza JA et al (2015) Pathogenesis and Transmission of Novel Highly Pathogenic Avian Influenza H5N2 and
H5N8 Viruses in Ferrets and Mice. Journal of virology 89: 10286-10293
influenza A and B viruses. Ibid. 85: 12825-12829
1511 Dengler L et al (2012) Immunization with live virus vaccine protects highly susceptible DBA/2J mice from lethal influenza
A H1N1 infection. Virology journal 9: 212
1512 Kim JI et al (2013) DBA/2 mouse as an animal model for anti-influenza drug efficacy evaluation. Journal of microbiology
(Seoul, Korea) 51: 866-871
1513 van der Vries E et al (2013) Prolonged influenza virus shedding and emergence of antiviral resistance in
immunocompromised patients and ferrets. PLoS pathogens 9: e1003343
1514 Belser JA et al (2011) The ferret as a model organism to study influenza A virus infection. Disease models & mechanisms 4:
575-579
influenza A and B viruses. Journal of virology 85: 12825-12829
1516 Dengler L et al (2012) Immunization with live virus vaccine protects highly susceptible DBA/2J mice from lethal influenza
A H1N1 infection. Virology journal 9: 212
1517 Tharakaraman K et al (2015) A broadly neutralizing human monoclonal antibody is effective against H7N9. Proceedings of
the National Academy of Sciences of the United States of America 112: 10890-10895
1518 Boon AC et al (2009) Host genetic variation affects resistance to infection with a highly pathogenic H5N1 influenza A virus
in mice. Journal of virology 83: 10417-10426
1519 Srivastava B et al (2009) Host genetic background strongly influences the response to influenza a virus infections. PloS one
4: e4857

mouse model system. 1520 Ferrets are naturally susceptible to a broader range of wild type influenza
viruses. The strength of this approach is that the use of wild type viruses and wild type hosts is more
relevant to human disease than other model systems. However, wild type viruses may display a limited
range of virulence, which limits their utility for this purpose. Moreover, the high genetic diversity among
influenza viruses complicates the comparison of results from the use of two genetically diverse wild type
strains that exhibit varying levels of pathogenicity.
observed during human infections are essential for the study of influenza-associated morbidity/mortality
and for testing the safety and efficacy of new vaccines and therapeutics. The strengths and limitations of
model systems that can be used to study influenza virus infection are summarized in Table 15.23.
Although the ability to infect wild type hosts with wild type viruses would be ideal for broad translation
of results to human populations, mice are naturally resistant to infection with many influenza viruses
and/or wild type viruses may display a limited spectrum of disease in mice and ferrets. In these cases,
because pathogenicity and disease outcome is dependent on the interplay between virus and host, both
GoF and alt-GoF approaches enable the development of model systems that expand the dynamic range of
pathogenesis that is observed when using wild type viruses and wild type hosts. GoF approaches achieve
this goal by enhancing the virulence of the virus through serial passaging, while alt-GoF approaches
enhance host susceptibility to disease through targeted genetic modification or the use of
immunosuppressants. Both approaches provide a controlled system for comparing the effectiveness of
MCM candidates to protect against more severe disease, and both have limitations. Serial passaging
(GoF) may change the phenotypic properties of the virus in ways that alter its susceptibility to the MCM
in development, which would lead to misrepresentative findings. Modification of the host (alt-GoF) may
alter host immune responses that are involved in the mechanism of action of the vaccine or therapeutic,
complicating translation of findings to disease in healthy hosts. The genetic modification approach is
limited to mice, although the use of immunosuppressants represents a promising approach for ferrets,
which are better representative of human disease. Given these caveats, the use of model systems derived
from GoF and alt-GoF approaches strengthens the validity of any findings.
1520 Margine I, Krammer F (2014) Animal models for influenza viruses: implications for universal vaccine development.
Pathogens (Basel, Switzerland) 3: 845-874.

Benefits to scientific knowledge and MCM development – Develop animal models for the study of influenza pathogenesis and to support MCM
development
• Generate influenza viruses that efficiently infect wild

GoF: type mice for the study of influenza pathogenesis and to • Adaptive mutations may alter the biology of
Adapt virus to host: serially passage virus support MCM development the virus and/or may alter susceptibility of the
in host (mouse or ferret) to increase • Generate genetically similar viruses that exhibit a large virus to MCMs, relative to the wild type virus
infectivity and virulence in that host range of virulence in ferrets, which provides a controlled
system for evaluating the efficacy of MCM candidates to
protect against severe disease
• Results in immunocompromised hosts may not
Alt-GoF #1: • Enables the study of wild type viruses translate to healthy populations
Sensitize host to influenza virus • Generate genetically similar hosts that exhibit a large • Limited to mice, which are less representative
infection: use of inbred mouse lines or range of disease severity in response to infection, which of human disease than ferrets
targeted knockout/transgenic mice provides a controlled system for evaluating the efficacy o There are limited tools for genetic
of MCM candidates to protect against severe disease modification of ferrets
Alt-GoF #2: • Enables the study of wild type viruses

Sensitize host to influenza virus • Generate genetically similar hosts that exhibit a large • Results in immunocompromised hosts may not
infection: treat host with range of disease severity in response to infection, which translate to healthy populations
immunosuppressants provides a controlled system for evaluating the efficacy
of MCM candidates to protect against severe disease
• Mice are resistant to infection with many
influenza viruses
Alt-GoF #3: • Wild type viruses may display a limited range
Infection of wild type hosts with wild
• Results using wild type viruses and wild type hosts are of virulence in naturally susceptible hosts
most likely to broadly translate to human populations
type viruses • Genetic diversity between wild type isolates
with naturally varying levels virulence
complicates comparison of results

15.4.4 Benefits of GoF to Surveillance
One major goal of influenza surveillance is to monitor the evolution of circulating animal influenza
viruses, in order to identify those viruses that pose a risk of emerging in human populations to cause a
pandemic. Resources can then be dedicated to mitigating the risk factors associated with virus emergence
and to preparing for a potential emergence event. Multiple virus properties contribute to the likelihood
that the virus will adapt to efficiently transmit in human populations and the potential consequences of
that event, including whether the virus is adapted (or poised to adapt) to efficiently infect and transmit in
humans, whether the population has pre-existing immunity to the virus, and viral virulence. As a result,
monitoring the virulence of circulating animal influenza viruses is one of the key goals of surveillance.
The strategies for monitoring the virulence of surveillance viruses are similar to those for monitoring
mammalian adaptation and transmissibility, and GoF approaches that enhance virulence and those that
enhance infectivity and transmissibility in representative animal models benefit surveillance through
similar mechanisms. Thus, these benefits are discussed collectively in section 15.3.4.
15.4.5 Benefits of GoF to the Development of Vaccines and Therapeutics
15.4.5.1 Vaccine Development Benefit #1: Development of New Influenza Vaccine Candidates
Standard methods for production of seasonal influenza vaccines have posed challenges for the production
of vaccines targeting highly pathogenic avian influenza strains such as H5N1, in part because the wild
type HPAI viruses are lethal to embryonated eggs, the main medium used for influenza vaccine
production.1521 In addition, egg-based production systems are not amenable to rapid scale-up due to their
reliance on the egg supply, which would pose a major problem if a novel pandemic virus emerged off
production cycle. For these reasons, researchers are exploring a variety of other platforms for the
production of vaccines for avian influenza viruses with pandemic potential. GoF approaches that enhance
virulence benefit the production of one of these platforms, live attenuated influenza vaccines (LAIVs).
Live attenuated influenza vaccines (LAIVs) are an attractive platform for pandemic vaccines for several
reasons: (1) the route of administration mimics the route of natural infection to trigger the generation of
mucosal and cell-mediated immunity, which is difficult to generate but is important for achieving robust
and long-term protection against mucosal pathogens such as influenza, and (2) LAIVs are quicker and
cheaper to manufacture than inactivated vaccines due to higher yields per egg and the fact that
inactivation and protein purification steps are not required, and (3) LAIVs can be easily administered via
intranasal drops or spray.1522 The major concern associated with LAIVs is their potential to regain
virulence in people, through reversion or the acquisition of compensatory mutations.1523 For that reason,
the WHO recommends serial passaging of LAIV candidates during the non-clinical phase of in vivo
toxicity and safety testing, to determine whether the LAIV is genetically stable or recovers virulence upon
1522 Ibid.
1523 Ibid.

passage in animals.1524, 1525 In accordance with these recommendations, multiple candidate LAIVs have
been subjected to serial passaging in animals.1526,1527,1528, 1529
15.4.5.1.2 Benefits and limitations of alt-GoF approaches
There are no alternative approaches that can provide similar information on the safety of LAIV
candidates.
Several alternative vaccine platforms which do not rely on GoF for their development, such as
recombinant vaccines, are also being explored. These vaccine platforms have strengths and limitations
relative to LAIVs (GoF). For example, adjuvanted, inactivated vaccines may provide broad-spectrum
immunity but require multiple doses to confer high levels of protection.1530
A variety of vaccine platforms are being explored for the development of vaccines targeting avian
influenza viruses with pandemic potential. LAIVs have several characteristics that are desirable for
pandemic vaccines, but a major concern associated with their use is that the LAIV may recover virulence
upon growth in people. GoF approaches are uniquely capable of demonstrating whether LAIV strains
recover virulence upon growth in vivo, a critical aspect of vaccine safety testing prior to the conduct of
clinical trials. Other types of vaccines in development have strengths and weaknesses relative to LAIVs.
The type or types of vaccines that will ultimately prove to be most effective for avian influenza viruses is
not yet clear based on vaccinology research conducted to date. Given the need for effective pandemic
influenza vaccines, pursuing all promising strategies for vaccine development in tandem, including
LAIVs, will ensure that an effective vaccine is achieved in the shortest possible period of time.
15.4.5.2 Vaccine Development Benefit #2: Targeted Mutagenesis to Remove Virulence Markers from
Vaccine Viruses
Most seasonal influenza vaccines are derived from whole vaccine viruses that are produced in eggs.1531
Existing production systems may also be used for the production of pre-pandemic vaccines and will be
used for the production of pandemic vaccines in response to the emergence of a novel pandemic strain.
The development of vaccines based on highly pathogenic avian influenza (HPAI) viruses such as H5N1
presents several challenges: (1) wild type viruses must be handled under biosafety level 3 (BSL-3)
containment due to their high pathogenicity, and (2) wild type viruses are lethal in chick embryos, the
medium used for production of most influenza vaccines in the US.1532 Thus, these viruses must be
attenuated in order to be safely and efficiently propagated in eggs for vaccine production. The multibasic
cleavage site in the influenza HA protein is a major determinant of viral virulence in eggs and chickens.
1524 WHO Expert Committee on Biological Standardization. (2010) Recommendations to assure the quality, safety and efficacy
of influenza vaccines (human, live attenuated) for intranasal administration. WHO Technical Report Series No 977, 2013.
The World Health Organization,, Geneva, Switzerland pp. 163-196.
1525 The World Health Organization. (2005) WHO guidelines on nonclinical evaluation of vaccines. WHO Technical Report
Series, No 927, 2005, Geneva, Switzerland, pp. 32-63.
1526 Jang YH, Seong BL (2012) Principles underlying rational design of live attenuated influenza vaccines. Clinical and
experimental vaccine research 1: 35-49
1527 Han P-F et al (2015) H5N1 influenza A virus with K193E and G225E double mutations in haemagglutinin is attenuated and
immunogenic in mice. Journal of General Virology 96: 2522-2530
1529 Sedova ES et al (2012) Recombinant influenza vaccines. Acta Naturae 4: 17-27
1532 Suguitan AL et al (2006) Live, Attenuated Influenza A H5N1 Candidate Vaccines Provide Broad Cross-Protection in Mice
and Ferrets. PLoS Med 3: e360

In addition to attenuating HPAI viruses through reassortment with attenuated, high-yield vaccine
backbone strains (e.g. PR8), as is standard for the production of all influenza vaccines in eggs, vaccine
manufacturers typically remove the HA multibasic cleavage site through targeted mutagenesis to further
attenuate the vaccine virus, enabling safe and efficient production of vaccine in eggs (or cells) under
BSL-2 conditions. In the future, other conserved determinants of virulence in the HA and NA proteins of
avian influenza (AI) viruses could be similarly deleted from AI vaccine viruses in order to further
improve the safety of the vaccine production process.
15.4.5.2.1 GoF approaches – benefits and limitations
As discussed above (section 16.4.2), GoF approaches, in particular forward genetic screens and serial
passaging, represent efficient and effective methods for discovering novel viral genetic and phenotypic
traits that contribute to virulence. This information provides a foundation for follow-up LoF studies that
aim to determine how to attenuate virulence, the goal of vaccine virus development, through mutation or
deletion of those traits.
15.4.5.2.2 Alt-GoF approaches – benefits and limitations
Several alt-GoF approaches can be used to discover novel virulence factors, including comparative
analysis of surveillance data, comparative analysis of the sequences of wild type viruses with varying
levels of virulence, use of replication incompetent viruses, and LoF forward genetic screens. As discussed
above, each of these approaches has critical limitations for the discovery of novel virulence traits relative
to GoF approaches. Namely, comparative sequence analysis and LoF screens are practically limited to the
investigation of known virulence traits due to the high genetic diversity among influenza viruses and the
inefficiency of screening mutants for attenuated virulence, respectively, and the relevance of novel traits
identified using in vitro replication-incompetent systems in the context of the complex host environment
is unknown.
However, following the identification of novel genetic traits that contribute to virulence, targeted
mutagenesis can be used to identify particular mutations within that genetic region that lead to attenuated
virulence. This LoF approach can also be used to demonstrate that the attenuating effect of particular
mutations is conserved in other virus strains.
The strengths and limitations of GoF and alt-GoF approaches for the discovery of virulence traits that can
be eliminated from vaccine viruses to improve the safety and efficacy of the vaccine production process
are summarized in Table 15.24. GoF approaches represent the most efficient and effective strategies for
discovery novel genetic traits that contribute to the virulence of influenza viruses. However, GoF
approaches cannot be used to identify or confirm genetic changes that are sufficient to attenuate the
virulence of wild type strains, which is the goal of vaccine virus development. LoF approaches, namely
targeted mutagenesis, are uniquely capable of identifying genetic changes (mutations or deletions)
attenuate virulence, as well as demonstrating that the attenuating consequences of those mutations are
conserved across multiple virus strains. Taken together, these approaches may enable the development of
novel virulence traits that can be mutated to attenuate virulence, which can be applied to the production of
AI vaccine viruses to further improve the safety of the vaccine production process.

Benefits to vaccine development – Targeted Mutagenesis to Remove Virulence Markers from Vaccine
Viruses
GoF Experimental Approaches:
• Serial passaging of viruses in • Most efficient and effective • Cannot be used to demonstrate
cells or animals strategies for discovering novel that deletion or mutation of a
virulence traits virulence trait is sufficient to
• Genetic modification to introduce
genetic traits expected to enhance attenuate virulence
virulence
Alt-GoF Experimental • Targeted LoF can be used to

Approaches: demonstrate that deletion or
• Genetic modification to introduce
mutation of a virulence trait is • Approaches are less efficient
sufficient to attenuate virulence and effective for the discovery
traits expected to attenuate
across multiple virus strains of novel virulence traits than
virulence (loss of function)
o Goal of application of GoF approaches
• Other approaches (see table knowledge to vaccine
15.20) development
15.4.5.3 Therapeutic development benefit #1: Inform the development of next-gen therapeutics
Only two classes of FDA-approved antivirals are approved for use in the U.S.: the adamantanes, which
inhibit the viral M2 protein, and the neuramidase inhibitors (NAIs), which include zanamivir (Relenza),
oseltamivir (Tamiflu), and peramivir (Rapivab).1533 The adamantanes are no longer recommended for use
due to widespread resistance.1534 Single mutations are sufficient to confer resistance to one or multiple
NAIs and have been observed in nature, though NAI-resistance mutations are not yet widespread.1535
Moreover, the NAIs exhibit limited efficacy, especially if administered more than 48 hours after symptom
onset.1536 Thus, there is an urgent need for the development of new therapeutics against influenza
viruses.1537 Researchers are actively working to develop next-gen influenza therapeutics that directly
target viral proteins as well as therapeutics that inhibit host factors that are critical for viral virulence or
that exacerbate infection-associated pathology. GoF approaches have potential to benefit the development
of both types of therapeutics.
15.4.5.3.1 Potential benefits and limitations of GoF approaches
As discussed in detail in section 15.4.3.1, GoF approaches represent the most efficient and effective
strategies for discovering novel viral genetic traits that contribute to pathogenicity, which may be good
targets for novel therapeutics. In addition, targeted genetic modification of viruses to introduce traits

1534 Ibid.
1535 Ibid.
1536 CDC. Use of Antivirals. Background and Guidance on the Use of Influenza Antiviral Agents.
http://www.cdc.gov/flu/professionals/antivirals/antiviral-use-influenza.htm. Last Update February 25, 2015. Accessed
November 28, 2015.

necessary and sufficient to enhance virulence across multiple virus contexts, which provides a strong
mechanistic basis for the role of that viral factor in virulence.
GoF approaches also enable the identification of host factors that are associated with virulence and
immunopathology, which may be good targets for novel host-targeted therapeutics. However, alt-GoF
approaches are needed to confirm the role of a particular host protein in virulence/immunopathology,
which provides an important conceptual foundation for the design of therapeutics targeting that protein.
Nonetheless, targeted modification to introduce mutations that are expected to enhance pathogenicity
(GoF) provides a controlled system for studying the interplay between virus and host factors that
contribute to pathogenicity, which is a valuable complement to alt-GoF approaches that perturb the
function of host factors, a more blunt approach.
Notably, in both cases, whether inhibiting viral or host factors discovered through GoF approaches is
sufficient to attenuate viral replication or infection-associated pathology must be empirically determined
using alt-GoF approaches.
15.4.5.3.2 Potential benefits and limitations of alt-GoF approaches
As discussed in detail in section 15.4.3.1.3, alt-GoF approaches have significant limitations for the
discovery of novel viral genetic traits and factors that contribute to virulence. In brief, unless genetically
similar viruses are available, approaches that rely on analysis of wild type viruses are limited to the study
of traits that are known to be associated with virulence, due to the high genetic diversity among influenza
viruses. Comparative analysis of isolates within patients enables the identification of novel adaptive traits
that are associated with enhanced virulence over the course of infection, but results may not be broadly
conserved in human populations. LoF screens are inefficient and may uncover traits that indirectly
contribute to pathogenicity. The use of replication-incompetent viruses enables the identification of novel
viral factors that contribute to viral fitness in vitro, but the importance of those genetic traits in the context
of the complex host environment is difficult to extrapolate. Finally, in vitro, virus-free approaches are
limited to the study of known virulence traits. However, alt-GoF approaches play a critical role in
investigating the function of putative virulence trait, to complement mechanistic information that can be
gleaned through GoF approaches. In particular, targeted LoF to confirm that blocking or attenuating the
function of a virulence factor attenuates viral replication and/or infection-associated pathology establishes
an evidence base for efforts to design therapeutics targeting that virulence factor.
Alt-GoF approaches provide valuable insight into host factors that enhance pathogenicity and contribute
to deleterious immune responses. Specifically, the use of targeted knockout animals or pharmacological
inhibition of the host factor during infection is uniquely capable of confirming that a host factor
contributes to virulence and pathogenicity. These approaches have been extensively used to discover host
factors that may be good targets for influenza therapeutics, including inhibitors of the NF-κB signaling
pathway, which enhances viral replication through several mechanisms, molecules that suppress levels of
reactive oxygen species, and inhibitors of cytokines/chemokines.1538,1539,1540,1541,1542 Because inhibition of a
1538 Wurzer WJ et al (2004) NF-kappaB-dependent induction of tumor necrosis factor-related apoptosis-inducing ligand
(TRAIL) and Fas/FasL is crucial for efficient influenza virus propagation. The Journal of biological chemistry 279: 30931-
30937
1539 Strengert M et al (2014) Mucosal reactive oxygen species are required for antiviral response: role of Duox in influenza a
virus infection. Antioxidants & redox signaling 20: 2695-2709
1540 Zhao D et al (2012) Phylogenetic and Pathogenic Analyses of Avian Influenza A H5N1 Viruses Isolated from Poultry in
Vietnam. PloS one 7: e50959
1541 Kash JC et al (2014) Treatment with the reactive oxygen species scavenger EUK-207 reduces lung damage and increases
survival during 1918 influenza virus infection in mice. Free radical biology & medicine 67: 235-247
1542 McKinstry KK et al (2009) IL-10 deficiency unleashes an influenza-specific Th17 response and enhances survival against
high-dose challenge. Journal of immunology 182: 7353-7363

host factor is likely to have multi-faceted effects on the immune response during infection, resolving the
function of host traits in viral clearance from deleterious immune responses may be difficult using this
approach. As a result, other alt-GoF approaches may be used to gain further mechanistic insight into the
role of the host factor during infection, including characterization of host immune responses to identify
host genes that are up-regulated during infection and LoF targeted genetic modification of viruses to tease
apart the role of particular virus-host interactions in pathogenesis.1543 Because comparative sequence
analysis provides minimal mechanistic insight and is of limited utility for discovering novel host factors
that contribute to pathogenicity, this alt-GoF approach does not contribute the design of new host-targeted
therapeutics.
In addition to designing therapeutics targeting specific virulence factors or pathways (virus or host),
several alternative strategies are used to develop novel candidate therapeutics. One alternative approach
for designing new therapeutics involves high-throughput screening of small molecule compounds to
identify compounds that reduce viral replication in vitro, which may identify candidate therapeutics that
target viral or host proteins.1544,1545 This approach has generated promising candidates, including
therapeutics that are in Phase III clinical trials in the U.S.1546 One drawback of this approach is that it is
limited to the identification of compounds that reduce viral replication, which is only one aspect of
virulence. Targeting other aspects of virulence, such as viral interactions with the host immune system,
may prove to be a more effective therapeutic strategy.
Another alternative approach involves identifying neutralizing monoclonal antibodies (mAbs) targeting
virus proteins. These approaches isolating mAbs that bind to particular virus proteins, such as the HA
protein, the nucleoprotein (NP), the NA protein, and the M2 protein, from the B cells of convalescent
patients or of mice that have been injected with the virus protein of interest.1547,1548,1549,1550,1551
Subsequently, the ability of mAbs to neutralize virus activity is tested. This approach has also generated
promising therapeutic candidates, including therapeutics that have entered Phase I clinical trials.1552,1553
However, mAb-based therapeutics have several drawbacks, including high production costs and the need
for injection-based delivery.1554
1543 Cheung CY et al (2002) Induction of proinflammatory cytokines in human macrophages by influenza A (H5N1) viruses: a
mechanism for the unusual severity of human disease? Lancet 360: 1831-1837
1544 Furuta Y et al (2002) In vitro and in vivo activities of anti-influenza virus compound T-705. Antimicrobial agents and
chemotherapy 46: 977-981
1545 An L et al (2014) Screening and identification of inhibitors against influenza A virus from a US drug collection of 1280
drugs. Antiviral research 109: 54-63
1546 Toyama Chemical Company, Ltd. Pipeline. https://www.toyama-chemical.co.jp/en/rd/pipeline/index.html. Last Update
1547 Krause JC et al (2011a) A broadly neutralizing human monoclonal antibody that recognizes a conserved, novel epitope on
the globular head of the influenza H1N1 virus hemagglutinin. Journal of virology 85: 10905-10908
1548 Clementi N et al (2011) A human monoclonal antibody with neutralizing activity against highly divergent influenza
subtypes. PloS one 6: e28001
1549 Bodewes R et al (2013) In vitro assessment of the immunological significance of a human monoclonal antibody directed to
the influenza a virus nucleoprotein. Clinical and vaccine immunology : CVI 20: 1333-1337
1550 Shoji Y et al (2011) An influenza N1 neuraminidase-specific monoclonal antibody with broad neuraminidase inhibition
activity against H5N1 HPAI viruses. Human vaccines 7 Suppl: 199-204
1551 Grandea AG, 3rd et al (2010) Human antibodies reveal a protective epitope that is highly conserved among human and
nonhuman influenza A viruses. Proceedings of the National Academy of Sciences of the United States of America 107:
12658-12663
1553 Visterra Pipeline. http://www.visterrainc.com/pipeline/pipeline.html. Last Update Accessed November 8, 2015.

The strengths and limitations of GoF and alt-GoF approaches for the development of new therapeutic
candidates are summarized in Table 15.25, below. GoF approaches represent the most efficient and
effective strategy for discovering novel viral virulence factors that may be good therapeutic targets, but
follow-up alt-GoF approaches are needed to confirm that inhibiting the function of a particular viral factor
is sufficient to attenuate or block viral replication and/or infection-associated pathology. Alt-GoF
approaches are best-suited for discovering novel host factors that contribute to virulence and
immunopathology. However, GoF approaches can be used to gain further mechanistic insight into the
function of the host protein during infection, which strengthens the evidence base for developing new
therapeutics targeting that host factor. Two completely different approaches for generating new
therapeutic candidates are screening libraries of small molecule compounds for their ability to inhibit viral
replication in vitro and isolating monoclonal antibodies that neutralize essential virus activities by directly
binding to virus proteins, both of which have generated promising therapeutic candidates that have
entered clinical trials. Given that influenza viruses readily acquire mutations that confer resistance to
therapeutics and that different types of therapeutics may be most effective against various influenza sub-
types, a wide repertoire of therapeutics is needed to best protect the public against the range of influenza
threats that exist in nature. Pursuing all promising pathways for therapeutic development in tandem,
including GoF approaches, is the best strategy to achieve this goal.

Benefits to therapeutic development – Develop new candidate therapeutics
GoF #1: • Most efficient and effective • Cannot demonstrate that inhibition
GoF Experimental Approaches: strategies for discovering of a given virulence factor is
novel viral virulence traits that sufficient to attenuate pathogenesis
• Serial passaging of viruses in are conserved across multiple
cells or animals virus strains, which may be • Limited utility for the discovery of
• Genetic modification to good targets for new novel host factors that contribute to
introduce genetic traits therapeutics virulence, relative to alt-GoF
expected to enhance virulence approaches
Alt-GoF #1: • Most efficient and effective

Alternative Experimental strategies for discovering • Results in immunocompromised
Approaches: novel host factors that hosts may not translate to healthy
contribute to virulence, which
• Genetic modification to may be good targets for new
populations
introduce traits expected to • Limited utility for the discovery of
therapeutics
attenuate virulence (loss of novel viral factors that contribute to
function) • Can be used to demonstrate virulence, relative to GoF
that blocking or attenuating
• Host-focused approaches the function of a viral
approaches
• Other approaches (see Table virulence trait is sufficient to
15.20) attenuate pathogenesis
Alt-GoF #2:
• Approach has generated • Limited to the discovery of
High-throughput screening of several promising therapeutic compounds that inhibit viral
small molecule compounds to candidates replication, which is only one aspect
identify those that inhibit viral
of pathogenesis
replication in vitro
Alt-GoF #3: • Approach has generated • mAb-based therapeutics have
Identify neutralizing monoclonal several promising therapeutic several drawbacks, including high
antibodies (mAbs) targeting candidates production costs and the need for
particular virus proteins injection-based delivery
15.4.6 Benefits to decision-making in public health policy
Evaluation of the virulence of circulating animal influenza viruses detected through surveillance informs
assessment of their pandemic risk, which informs prioritization of investments in pre-pandemic
preparedness initiatives, such as pre-pandemic vaccine development. This GoF benefit to decision-
making in public health policy is discussed in detail in section 15.3.5, as evaluation of the transmissibility
of animal influenza viruses similarly informs pandemic risk assessments and downstream decision-
making.

15.5 Influenza viruses: Detailed analysis of the benefits of GoF research that leads to
evasion of existing natural or induced adaptive immunity
15.5.1 Overview of the Influenza GoF Landscape
lead to evasion of existing natural or induced adaptive immunity. In this section, an overview of GoF
approaches in this phenotypic category and describe the scientific outcomes and/or products of each
approach.
15.5.1.1 Serial passaging of viruses in the presence of cognate antibodies
Serial passaging of viruses in the presence of cognate antibodies may lead to the acquisition of mutations
that allow the virus to escape neutralization by the antibody. This experiment can be performed in cell
culture using monoclonal antibodies, convalescent sera from infected individuals, post-infection ferret
sera, or in animals that have been vaccinated or previously exposed to influenza viruses. Sequencing of
emergent antibody escape viruses identifies amino acid substitutions that are sufficient to confer antigenic
change, which provides a foundation for follow-up studies investigating the molecular basis of antigenic
differences between strains. Additionally, sequencing viral isolates at multiple stages of the selection
process and determining the effect of amino acid substitutions on viral fitness and other virus phenotypes
provides insight into the evolutionary mechanisms driving antigenic drift. Finally, when performed in
vitro using monoclonal antibodies, the location of escape mutations reveals potential antibody epitope
sites.
15.5.1.2 Forward genetic screen to identify mutations that alter antigenicity
Forward genetic screens involve random mutagenesis of the HA protein followed by characterization of
the antigenicity of mutants using the hemagglutination inhibition (HAI) assay or other assays, in order to
identify amino acid substitutions that do and do not lead to antigenic change. Follow-up studies may
determine the consequences of antigenicity-altering mutations on other virus phenotypes, such as viral
fitness and pathogenicity. As for serial passaging experiments, the identification of amino acid
substitutions that confer antigenic change provides a foundation for studies investigating the molecular
basis of antigenic differences. In addition, comprehensive forward genetic screens can be used to define
the ‘antigenic landscape’ of the HA protein – that is, which substitutions the HA protein will tolerate and
which of those substitutions cause antigenic drift.
15.5.1.3 Targeted modification of viruses to introduce mutations that are expected to alter antigenicity
A final GoF approach that may lead to viruses that evade existing adaptive immunity involves targeted
genetic modification to introduce mutations that are expected to alter antigenicity, followed by antigenic
characterization of the mutant virus using the HAI assay or other assays. Of note, mutations may be
identified through GoF approaches, such as serial passaging of viruses in the presence of cognate
antibodies, or alt-GoF approaches, such as comparative analysis of historical sequences. This approach
demonstrates that a particular mutation or set of mutations is necessary and sufficient to alter antigenicity,
which provides a foundation for follow-up studies investigating the molecular basis of antigenic
differences between strains.
When any of these three approaches are performed using recently or currently circulating seasonal
influenza viruses or using seasonal viruses that have recently served as the basis for vaccine strains, the
end result is the generation of a mutant strain that cannot be neutralized by existing natural or induced
immunity, respectively. Because human populations do not have widespread immunity to the 1918 H1N1

pandemic virus or to animal influenza viruses (i.e. avian viruses1555 and swine viruses1556), no approaches
involving these viruses meet this phenotypic criterion. Therefore, this section does not include studies that
investigate the mechanisms underlying antigenic drift of avian strains in response to selection pressure
from vaccination or the chicken immune system, nor any other studies focused on animal influenza
strains. Note that because these studies may lead to the acquisition of mutations in the influenza HA
protein, which is a critical determinant of mammalian adaptation, transmissibility, and virulence, these
studies may result in the generation of viruses with altered virulence, infectivity, and transmissibility from
a “human” perspective. However, whether and what phenotypic changes are likely to arise cannot be
anticipated with certainty.
Finally, GoF approaches may also lead to the generation of influenza viruses that are capable of evading
recognition by the host innate immune system. Because virus interactions with innate immune factors are
critical determinants of virulence, these approaches are evaluated in section 15.4.
15.5.2 Overview of the potential benefits of GoF experiments that may lead to the generation of
influenza viruses that evade existing natural or induced adaptive immunity
GoF approaches have the potential to benefit several aspects of scientific knowledge about the antigenic
drift of influenza viruses. First, GoF studies that identify mutations that confer antigenic change provide a
foundation for follow-up studies investigating the molecular basis of antigenic differences between
strains. Second, GoF studies provide insight into the evolutionary mechanisms driving antigenic drift in
response to immune pressure. Finally, GoF studies enable the identification of antigenic sites on the HA
protein, which can also provide insight into both aspects of antigenic drift described above.
GoF approaches that lead to the identification of mutations that alter antigenicity have potential to inform
the interpretation of surveillance data for human seasonal influenza by facilitating inference of antigenic
phenotype from genotype, in lieu of isolating and antigenically characterizing viruses. Specifically, these
data inform the development of models for predicting antigenic phenotype from genotype, or surveillance
sequences can be examined for the presence of absence of particular amino acid substitutions that are
associated with antigenic change. Either application has the potential to inform the bi-annual selection of
strains for the seasonal influenza vaccine.
Because this GoF phenotype is restricted to the study of human seasonal influenza viruses, GoF
approaches in this category do not benefit surveillance in wildlife or agricultural animals.
15.5.2.3 Vaccines
GoF approaches have potential to improve the strain selection process for seasonal influenza vaccines in
several ways. First, a critical factor in strain selection is analysis of the antigenic characteristics of
circulating influenza viruses, to determine whether new antigenic variants have emerged. As described in
section 15.5.2.2, GoF data may facilitate prediction of antigenic phenotype from genotype, which may
provide several advantages over the use of traditional, laboratory-based antigenic characterization
methods. In addition, GoF approaches have the potential to aid efforts to predict antigenic drift, either
1555 Jernigan DB, Cox NJ (2015) H7N9: Preparing for the Unexpected in Influenza. Annual Review of Medicine 66: 361-371
1556 Skowronski DM et al (2012) Cross-reactive and vaccine-induced antibody to an emerging swine-origin variant of influenza
A virus subtype H3N2 (H3N2v). J Infect Dis 206: 1852-1861

directly through the selection and analysis of drifted strains or by informing the development of models
for predicting drift. As selected strains sometimes drift during the course of vaccine development, which
leads to poor vaccine match, either effort could improve the efficacy of vaccines by enabling deliberate
production of “drifted” strains that match circulating strains at the time of vaccine deployment.
GoF approaches in this phenotypic category are focused on elucidating mechanisms of antigenic drift in
response to immune pressure, which is not relevant for the development of therapeutics. (We note that
studies that generate escape mutants from candidate monoclonal antibody therapeutics, which are
experimentally similar to approaches described above, are discussed in section 15.7.)
GoF approaches have potential to inform the selection of strains for the seasonal influenza vaccine in
several ways, as described in section 15.5.2.3.
15.5.2.6 Economic benefits
GoF approaches that inform strain selection for seasonal influenza vaccines may improve the efficacy of
seasonal flu vaccines by increasing the likelihood that the vaccine strains will match the strains that are
circulating during the target influenza season. Ultimately, this benefit may increase vaccine uptake but
otherwise is unlikely to yield economic benefits.
Influenza viruses circulating in nature acquire mutations in response to immune pressure from human
populations that allow the viruses to escape recognition by the adaptive immune system, a process termed
“antigenic drift”.1558 As a result, the strain composition of the seasonal influenza vaccine must be updated
annually to ensure that the vaccine strains antigenically “match” circulating strains. Research in this area
is focused on the influenza HA protein, which is the immunodominant influenza protein and represents
the primary component of current influenza vaccines. (The role of other influenza proteins, such as
neuraminidase, in the adaptive immune response is not well understood and is an active area of research.
Given this uncertainty, this section does not evaluate studies that investigate virus escape from antibodies
against non-HA influenza proteins.) The mechanisms underlying antigenic drift of the HA protein and the
relationship between genotype and antigenic phenotype are not well understood. One of the knowledge
gaps that contributes to this uncertainty is an incomplete understanding of the antigenic sites on the HA
protein that are targeted by human monoclonal antibodies, as these sites are presumably hotspots for
antigenic evolution.1559 Most work to map antibody epitopes has been conducted using murine antibodies,
which exhibit some distinctive antibody binding characteristics relative to human mAbs. Additionally,
although the major antigenic sites on the H1 protein were defined in the early 1980s using the lab-adapted
1558 Webster RG et al (1982) Molecular mechanisms of variation in influenza viruses. Nature 296: 115-121

A/Puerto Rico/8/1934 (PR8) strain, the antigenic regions of the H1 protein from the 2009 H1N1
pandemic strain may be different and have been the subject of several preliminary studies.1560,1561,1562,1563
(Experimentally mapping the antibody epitopes of future pandemic strains will be an important research
and public health goal if a new pandemic strain emerges.)
In this section, the ability of GoF methods, versus alternative approaches, to address three unanswered
questions in this field are evaluated:
• How do influenza viruses evolve antigenically in response to immune pressure? That is, what are
the evolutionary mechanisms driving antigenic drift, including the role of different selection
pressures (e.g. vaccination) and the interplay between antigenic escape and other virus
phenotypes, such as fitness?
• What is the molecular basis of antigenic drift? That is, what amino acid substitutions in the HA
protein lead to antigenic change, and what is the biophysical basis of that effect?
• What are the antigenic sites on the HA protein that are targeted by human monoclonal antibodies?
For each question in turn, the potential benefits and limitations of relevant GoF approaches and alt-GoF
approaches are described, then the benefits of GoF approaches relative to alt-GoF approaches are
evaluated. Unique benefits of GoF and alt-GoF approaches are highlighted.
15.5.3.1 Scientific Knowledge Gap 1 – How do influenza viruses evolve antigenically in response to
immune pressure?
GoF approaches that involve serial passaging of viruses in the presence of cognate antibodies provide
insight into the evolutionary mechanisms driving antigenic drift in response to immune pressure. Both in
vivo and in vitro approaches have unique strengths. Namely, subjecting viruses to selection from the full
complement of the animal immune system better mimics the selective pressure viruses experience in
humans, while in vitro approaches can be conducted using convalescent sera (or isolated antibodies) from
people, which may be more relevant to humans than selective pressures in animals. In addition, the in vivo
approach represents a controlled system for studying the role of selective pressures from prior exposure to
influenza viruses through natural infection and/or vaccination in shaping antigenic evolution. In both
cases, results from laboratory studies may not translate to the evolution of viruses in human populations
in nature and may not be conserved in other virus contexts. Importantly, follow-up studies can determine
the effect of antigenic drift on other virus phenotypes, such as fitness, which provides insight into how
likely mutations are to persist in a host or in a population once they have arisen.
1560 Caton AJ et al (1982) The antigenic structure of the influenza virus A/PR/8/34 hemagglutinin (H1 subtype). Cell 31: 417-
427
1562 Rudneva I et al (2012) Escape mutants of pandemic influenza A/H1N1 2009 virus: variations in antigenic specificity and
receptor affinity of the hemagglutinin. Virus Res 166: 61-67
1563 Krause JC et al (2011b) A Broadly Neutralizing Human Monoclonal Antibody That Recognizes a Conserved, Novel
Epitope on the Globular Head of the Influenza H1N1 Virus Hemagglutinin. J Virol 85: 10905-10908

The use of attenuated strains for serial passaging studies, in lieu of wild type strains, represents one type
of alt-GoF approach. Two types of attenuated strains are used for serial passaging studies to investigate
antigenic evolution mechanisms: the mouse-adapted strain PR8, which is avirulent in people,1564 and 6:2R
strains that contain the HA and NA gene segments from a seasonal strain of interest and the remaining six
gene segments from PR8. While use of either type of attenuated strain can provide insight into the basic
mechanisms of antigenic evolution, results may not translate to wildtype strains due to differences in
disease pathogenesis caused by wildtype versus attenuated strains and other factors. Another potential
concern is that relative HA and NA expression levels may be different in the context of a 6:2R, as the
effect of HA/NA balance on antigenic drift is as yet unknown. Moreover, 6:2R strains cannot be used to
predict the effect of antigenic escape mutations on the fitness of wildtype strains because in vivo fitness is
a complex, multi-genic trait that is highly dependent on genetic context. As the PR8 strain and 6:2R
strains do not efficiently infect ferrets,1565 these studies are limited to the use of mouse model systems.
Comparative analysis of historical virus sequences that have drifted antigenically over time represents an
alternative experimental approach for studying antigenic evolution. Relative to GoF approaches, the
strength of the comparative sequence analysis approach is that it provides insight into the antigenic
evolution of a wide breadth of influenza viruses in human populations. However, the success of this
approach depends on the quality of available surveillance data; some strains have limited numbers of
sequences available, and biases in the way that some surveillance data are collected render the data
unsuitable or difficult to use. Moreover, given the variability in levels of pre-existing immunity in the
population due to differences in infection and vaccination histories, inferring how selective pressures
from vaccination and/or prior infection shape antigenic drift may be difficult. An additional limitation is
that the historical record is static – that is, it cannot provide insight into mutations that were selected
against, which is important knowledge for understanding the pressures and constraints that guide
antigenic evolution. Furthermore, the extent of information that can be generated using this approach is
constrained by the fact that history has only explored a fraction of the possible antigenic space, for a
given influenza subtype. Finally, this approach cannot be used to proactively study the antigenic evolution
of currently circulating viruses.
In silico approaches can be also used to investigate mechanisms underlying antigenic drift of influenza
viruses. Existing models are largely based on historical sequence data and accompanying antigenic
characterization data and have been validated using historical data. As a result, the quality of these models
is constrained by the set of limitations described above for the comparative sequence analysis approach.
Although models can provide insight into the relationships between genetic and antigenic evolution, their
accuracy in predicting future antigenic drift is unknown, thus any predictions must be experimentally
validated. Additional experimental data about pathways for antigenic evolution, including data generated
using GoF approaches, is needed to improve the quality of existing models.
Table 15.26 summarizes the benefits and limitations of GoF and alt-GoF approaches that provide insight
into the antigenic evolution of influenza viruses. Taken together, GoF approaches are uniquely capable of
providing in-depth information about the evolutionary mechanisms driving antigenic drift as well as
prospective information about the evolution of currently circulating viruses. In vivo approaches provide
insight into antigenic drift in response to selective pressure from the full complement of the immune
1564 Beare AS et al (1975) Trials in man with live recombinants made from A/PR/8/34 (H0 N1) and wild H3 N2 influenza
viruses. Lancet 2: 729-732
1565 Jin H et al (2004) Imparting Temperature Sensitivity and Attenuation in Ferrets to A/Puerto Rico/8/34 Influenza Virus by
Transferring the Genetic Signature for Temperature Sensitivity from Cold-Adapted A/Ann Arbor/6/60. J Virol 78: 995-998

system but may not translate to humans, while in vitro approaches can provide information about
antigenic changes that arise in response to selective pressure from human antibodies but may not translate
to complex, in vivo scenarios. In either case, lessons learned in the laboratory may not translate to virus
behavior in human populations in nature. In contrast, comparative sequence analysis is uniquely capable
of providing information about the antigenic evolution of viruses in nature, but is constrained to reactively
studying the evolution of historic viruses in limited depth.

Table 15.26. Summary of the benefits of GoF approaches that lead to evasion of existing immunity
Scientific Knowledge Benefits – What are the evolutionary mechanisms underlying antigenic drift?
• Provide insight into evolutionary mechanisms driving

antigenic drift • Artificiality – adaptive changes observed in the
GoF #1 [1]*: laboratory may not be representative of
o Directly translates to humans, if convalescent sera evolution in nature
In vitro: Serial passaging of virus in the (or isolated antibodies) from people are used
presence of monoclonal antibodies for one or • Narrow breadth – results may not generalize to
more passages • Provide insight into the consequences of antigenic drift other virus contexts
for replicative fitness
• Translatability – in vitro results may not
• Proactive - can be performed using currently translate to humans
circulating viruses
• Provide in-depth insight into evolutionary mechanisms

driving antigenic drift
o Selective pressure from full complement of the
animal immune system mimics selective pressure in • Artificiality - adaptive changes observed in the
humans laboratory may not be representative of
GoF #2 [2]: o Identifies positively and negatively selected traits evolution in nature
In vivo: Serial passaging of virus in vaccinated o Controlled system for studying role of selective • Narrow breadth - results may not generalize to
animals or animals with prior exposure to pressure from prior exposure to influenza viruses other virus contexts
influenza viruses through vaccination and/or natural infection
• Translatability – results from animal models
• Provide insight into the consequences of antigenic drift may not translate to humans
for other viral phenotypes, such as fitness
• Proactive – can be performed using currently
circulating viruses

Scientific Knowledge Benefits – What are the evolutionary mechanisms underlying antigenic drift?
• Limited by the quality and availability of

• Provide insight into evolutionary mechanisms driving existing surveillance data
antigenic drift • Reactive – limited to the study of antigenic
Alt-GoF #1 [1]:
o Provides information on the natural evolutionary evolution that has already occurred in nature
Comparative analysis of historical virus process
sequences • Static – cannot identify lost or negatively
o Directly translates to humans selected traits
o Analyzes broad datasets applicable to many strains • Variability in levels of pre-existing immunity
in surveillance populations complicate
interpretation of selection pressures
• Predictive – does not confirm or correlate

phenotypic effects in a biological context
• Model accuracy – existing models are based on
Alt-GoF #2 [2]: • Gain insight into evolutionary mechanisms of antigenic historical data
In silico, virus free: Use computational or drift o Limited by quality and availability of
mathematical approaches to build models for • Proactive – can be applied to currently circulating existing surveillance data
prediction of future antigenic drift viruses o Accuracy in predicting future antigenic drift
is unknown
• Cannot predict consequences of antigenic drift
on other viral phenotypes
* GoF and alt-GoF approaches are listed in numerical order. Numbers in brackets specific approaches described in the landscape tables (Supplemental
information).

15.5.3.2 Scientific Knowledge Gap 2 – What is the molecular basis of antigenic drift?
Several GoF approaches can be used to discover mutations that lead to antigenic drift, which provides a
foundation for follow-up studies investigating the biophysical basis of antigenic change. First, serial
passaging of viruses in cells in the presence of cognate sera or monoclonal antibodies, or in animals that
have been vaccinated or previously exposed to influenza viruses, leads to the emergence of antigenic
escape mutants. Sequencing the HA gene of emergent escape viruses reveals mutations that are sufficient
to alter virus antigenicity. This approach is highly efficient and can be applied to any virus, including
currently circulating strains. Notably, in vitro and in vivo selection approaches equally enable the
identification of mutations associated with antigenic drift, though the in vitro approach is faster and
cheaper. Importantly, as multiple mutations may arise during passaging, follow-up studies may be needed
Forward genetic screens, which involve mutagenesis of the HA protein and subsequent characterization of
the antigenicity of mutant viruses, represent another GoF approach for identifying mutations that confer
antigenic change. Though screening for escape mutants is more labor-intensive than selection methods
based on serial passaging, the screening approach is uniquely capable of identifying mutations that do not
lead to antigenic change, which critically informs efforts to develop models for the sequence-based
prediction of antigenicity. In addition, comprehensive mutagenesis of the HA protein enables
characterization of the “antigenic landscape” of HA – that is, the set of amino acid substitutions that HA
can tolerate and the subset of those that lead to antigenic change. Understanding the antigenic plasticity of
the HA protein provides important context for evaluating the molecular basis of antigenic drift and may
benefit the development of new influenza vaccines, as described below. Importantly, because of the
influence of genetic context on antigenicity, antigenic escape mutations identified through either serial
passaging or forward genetic screens may not generalize to other virus strains within the same or different
HA subtype.
Finally, targeted genetic modification of viruses to introduce mutations associated with antigenic change,
followed by antigenic characterization of mutant viruses, is used to demonstrate that mutations are
necessary and sufficient to alter antigenicity. Notably, these mutations may be identified through GoF
approaches, such as serial passaging, or alt-GoF approaches, such as comparative sequence analysis
(described below). Subsequently, to determine whether the phenotypic consequences of mutations are
functionally generalizable across multiple virus strains, targeted mutagenesis can be used to introduce
mutations into new virus strains, followed by antigenic characterization. Together, these results provide a
strong foundation for follow-up structural studies to determine the biophysical basis of antigenic
differences and critically inform the development of models for the prediction of antigenic phenotype
from genotype.
Because experiments in this phenotypic category focus on the influenza HA protein, reassortment strains
containing the HA and NA genes from a seasonal strain of interest and the remaining six “internal” genes
from the lab-adapted, attenuated strain PR8 (6:2R strains) can be used in lieu of wildtype seasonal strains
for any of the GoF approaches described above. Due to the fact that a 6:2R strain is attenuated relative to
the parental HA/NA donor strain, use of 6:2R strains represents one type of alt-GoF approach. Because
the antigenicity of the HA protein is preserved in the context of a 6:2R strain,1566 6:2R strains are as
1566
(2015h) Interviews with influenza researchers.

suitable as wild type strains for the discovery and confirmation of amino acid substitutions that lead to
antigenic drift using in vitro or mouse model systems. Although 6:2R strains do not efficiently infect
ferrets, this limitation does not compromise the utility of 6:2R strains because ferret experiments do not
provide unique information about antigenic escape mutations relative to the use of other model systems.
Several alternative experimental approaches can also be used to identify mutations associated with
antigenic change. Comparatively analyzing the sequences of natural isolates that have drifted
antigenically over time can lead to the identification of mutations that are associated with antigenic
change. Even though the major antigenic sites on the HA protein have been mapped, not all mutations
within those sites cause antigenic changes and mutations outside those sites may lead to antigenic changes
through long-range effects. Current models cannot accurately predict which mutations do or do not lead to
antigenic drift, necessitating follow-up experiments to determine which of the identified mutation(s) lead
to antigenic drift. The key drawback of this approach is that it is limited to the identification of mutations
that have arisen in nature, which represents a fraction of the possible antigenic space.
In silico approaches represent another alt-GoF approach for the identification of mutations associated
with antigenic drift. Specifically, computational models based on antigenic, sequence, and HA structural
data can be used to predict amino acid substitutions that will alter antigenicity. Although computational
approaches can fully explore all possible antigenic configurations, existing models cannot predict
mutations that will lead to antigenic change with certainty, thus the phenotypic consequences of any
predicted mutation must be confirmed experimentally. Notably, because existing models are primarily
based on historical sequence data and accompanying antigenic characterization data, the quality of these
models is constrained by the set of limitations described above for the comparative sequence analysis
approach. Additional experimental data, including data generated from GoF experiments, is needed for
parameterization of improved models.1567
Table 15.27 summarizes the benefits and limitations of GoF and alt-GoF approaches that can provide
insight into the molecular basis of antigenic drift. Taken together, GoF approaches are uniquely capable
of identifying amino acid substitutions that are necessary and sufficient to alter antigenicity, which
provides a critical foundation for the follow-up studies to elucidate the biophysical basis of antigenic
differences. Furthermore, GoF approaches represent the most efficient and reliable method for uncovering
mutations that cause antigenic drift in circulating strains and are uniquely capable of exploring antigenic
space to define which mutations do and do not lead to antigenic changes, which can improve predictive
modeling efforts. For the purpose of discovering mutations that lead to antigenic change, GoF approaches
can be conducted using attenuated 6:2R strains instead of wildtype strains without compromising the
quality and accuracy of the information that is generated.
1567 (2015) Interviews with influenza researchers.

Table 15.27. Summary of the benefits of GoF approaches that lead to evasion of immunity to scientific knowledge
Scientific Knowledge Benefits – What is the molecular basis of antigenic drift?
GoF #1 [1]*:
• Discover novel mutations that are
In vitro: Serial passaging of virus in the presence of sufficient to confer antigenic change
monoclonal antibodies for one or more passages • Associative – information produced may be
o Gain insight into biophysical basis correlative, not causative
of antigenic differences
• Narrow breadth – results may not generalize to other
GoF #2 [2]: • Proactive – can be applied to any virus strains
In vivo: Serial passaging of virus in vaccinated virus strain, including currently
animals circulating strains
• Discover novel mutations that do and

do not cause antigenic drift
o Gain insight into biophysical basis
GoF #3 [3]: of antigenic differences
• Narrow breadth - results may not generalize to other
virus strains
Forward genetic screen to identify mutations that o Gain insight into the antigenic
alter antigenicity plasticity of the HA protein • Screening approach is more labor-intensive than
selection approaches
• Proactive - can be applied to any
virus strain, including currently
circulating strains
• Confirm that mutations are necessary

and sufficient to confer antigenic
change
GoF #4 [4,5]: • Lack of publication of negative results: Compromises
o Gain insight into biophysical basis
Targeted genetic modification to introduce mutations of antigenic differences evaluation of whether the function of particular
that are expected to alter antigenicity markers is broadly conserved
• Proactive - can be applied to any
virus strain, including currently
circulating strains

Table 15.27. Summary of the benefits of GoF approaches that lead to evasion of immunity to scientific knowledge
Scientific Knowledge Benefits – What is the molecular basis of antigenic drift?
• Associative – information produced is correlative, not

causative
Alt-GoF #1 [1]: • Identify mutations that are associated
Comparative analysis of historical sequences with altered antigenicity • Reactive - limited to the study of antigenic space that
has already been explored in nature
• Predictive – does not confirm or correlate phenotypic

effects in a biological context
• Predict novel mutations that may
Alt-GoF #2 [2]:
lead to antigenic changes • Model accuracy – existing models are based on
In silico, virus free: Use computational or historical data
mathematical approaches to build models for • Proactive - can be applied to any o Historical influenza viruses have explored a
prediction of antigenicity based on genotype virus strain, including currently
fraction of the possible antigenic space
circulating strains
o Accuracy in predicting phenotypic consequences of
novel mutations is unknown
information).

15.5.3.3 Scientific knowledge gap 3: What are the antigenic sites on the HA protein that are targeted by
human monoclonal antibodies?
Serial passaging of viruses in cells in the presence of monoclonal antibodies to select for antibody escape
mutants is a classic method for identifying putative antibody binding sites. Specifically, the amino acid
positions where mutations arise represent potential antigenic sites, although interpretation of this data is
complicated by the fact that mutations outside antibody binding sites can alter the confirmation of HA to
impact HA-antibody interactions through long-range effects. In the event that multiple mutations arise
within the HA protein, targeted mutagenesis to introduce individual mutations into the parental strain may
be used to confirm which mutations are necessary and sufficient to confer escape. Together, these
approaches can be used to map the epitope of a particular monoclonal antibody or to comprehensively
map antigenic sites through the use of multiple, distinct mAbs. In the latter case, a collection of escape
mutants is generated using several different mAbs, and subsequent testing of whether each escape mutant
can be neutralized by each mAb (i.e. all possible cross-reactions) reveals conserved and distinct
epitopes.1568,1569,1570 This approach is simple, rapid, and allows for precise mapping of antigenic sites.
However, each passaging experiment focuses on the identification of a single antigenic site, such that
multiple rounds of passaging with distinct antibodies are required to map multiple antigenic regions.
Several alt-GoF approaches can also be used to map the antigenic epitopes of the influenza HA protein.
One approach involves the use of cell surface display systems in yeast, bacteria, or bacteriophages. These
systems exploit the ability of these organisms to express random peptides or protein fragments from the
HA protein on their cell surface. Libraries of mutant bacteria/phages/yeast can then be screened for
binding to a monoclonal antibody or post-infection sera, for mapping of the antigenic epitope of a
particular antibody or comprehensive mapping of antigenic sites, respectively. The main strength of this
approach is that it is high-throughput, allowing for mapping of multiple antigenic sites at once through the
use of complex sera or multiple mAbs. However, as the presentation of mapped epitopes may be different
in the context of the full virus, experiments with full virus should be performed to validate any findings.
(We note that validation would entail determining whether mutagenesis of putative antibody binding sites
abrogates antibody neutralization, a GoF experiment.)
Another alternative approach involves analysis of crystal structures of a viral protein (or protein fragment)
complexed with a particular mAb. The crystal structure demonstrates precisely where an antibody binds
to the HA protein, which can be compared to previous studies to determine whether the epitope is
previously known or novel. The main drawback of this approach is that it is labor- and time-intensive and
therefore has limited throughput. Additionally, researchers have faced technical limitations, such as
difficulty crystallizing full-length HA proteins and radiation damage during the data collection process,
which may compromise the quality of the data. 1571
427
1569 Gerhard W et al (1981) Antigenic structure of influenza virus haemagglutinin defined by hybridoma antibodies. Nature 290:
713-717
1570 Matsuzaki Y et al (2014) Epitope mapping of the hemagglutinin molecule of A/(H1N1)pdm09 influenza virus by using
monoclonal antibody escape mutants. Journal of virology 88: 12364-12373
1571 Hong M et al (2013) Antibody Recognition of the Pandemic H1N1 Influenza Virus Hemagglutinin Receptor Binding Site.
Ibid. 87: 12471-12480

into the antigenic sites of the HA protein that are targeted by human monoclonal antibodies. Serial
passaging of viruses in the presence of antibodies, a GoF approach, represents the only method for
mapping the antigenic sites of the HA protein in the context of a full virus. However, the fact that
mutations outside of antigenic sites may confer escape through long-range effects complicates
interpretation of mutational data from these experiments. In addition, the approach is relatively low-
throughput in that each passaging experiment enables identification of a single antigenic site. In contrast,
the use of cell surface display systems in yeast, bacteria, or phages represents a high-throughput method
for identifying the antigenic sites of particular mAbs or for comprehensively mapping the antigenic sites
on a given HA protein. Analysis of the crystal structures of HA-antibody complexes precisely reveals the
antibody binding site, but the resources needed and technical challenges associated with this approach
render it low-throughput. Additionally, the relevance of both in vitro approaches is limited by the fact that
that HA presentation may differ in the context of the full virus. Confirming the results of an in vitro
experiment requires determining whether mutating the proposed antigenic sites allows for escape from
antibody neutralization, a GoF approach.

Scientific Knowledge Benefits – What are the antigenic epitopes on the HA protein?
• Associative – if more than one

mutation arises during
GoF #1 [1]*: • Discover putative antigenic passaging
In vitro: Serial passaging of virus in epitopes on the HA protein in • Mutations outside binding sites
the presence of monoclonal antibodies the context of the full virus may confer antigenic escape
for one or more passages through long-range effects
• Enables identification of a
single antigenic site
GoF #2 [4,5]:
• Mutations outside binding sites
• Confirm putative antigenic may confer antigenic escape
Targeted genetic modification to through long-range effects
epitopes on the HA protein in
introduce mutations expected to confer
the context of the full virus • Enables identification of a
antigenic escape
single antigenic site
• Discover putative antigenic

Alt-GoF #1 [3]: epitopes on the HA protein
In vitro, virus free Cell surface • High-throughput • Simplicity of model system –
expression of HA peptides or results may not be recapitulated
o Enables screening with
fragments in yeast/phages/bacteria in the context of the full virus
multiple mAbs or complex
• Screen library for antibody binding sera to map multiple
antigenic sites at once
• Simplicity of model system –

results may not be recapitulated
Alt-GoF #2 [4]:
• Discover antibody binding in the context of the full virus
In vitro, virus free Analysis of crystal
sites on HA proteins • Low-throughput – X-ray
structures of HA-antibody complexes
crystallography is labor-
intensive
* GoF and alt-GoF approaches are listed in numerical order. Numbers in brackets specific approaches described
in the landscape tables (Supplemental information).
15.5.4 GoF Benefits to Surveillance
15.5.4.1 Surveillance benefit #1: Aid in the interpretation of seasonal influenza genetic surveillance
data
The WHO Global Influenza Surveillance and Response System (GISRS) conducts surveillance of
seasonal influenza viruses year-round. The major goal of seasonal flu surveillance is to monitor the
antigenic evolution of viruses – that is, to detect when new antigenic variants emerge in human

populations and to determine their prevalence and geographic distribution.1572,1573 GISRS is a two-tiered
surveillance and public health laboratory system.1574,1575 A global network of National Influenza Centres
(NICs) collect clinical specimens in their countries, perform preliminary analyses such as viral isolation
and sub-typing, and forward representative virus isolates to one of six WHO Collaborating Centres
(WHOCCs) for further characterization. WHOCCs, which include the CDC and St. Jude Children’s
Research Hospital in the U.S., conduct antigenic characterization assays, sequencing, and several other
virus characterization assays. These data critically inform WHO-coordinated decisions about which
strains to recommend including in the seasonal flu vaccine, which are developed during bi-annual
Vaccine Composition Meetings (VCMs).1576,1577 If surveillance data indicate that a new antigenic variant
has emerged and spread geographically, the WHO strain selection committee will recommend updating
that component of the vaccine.
Antigenic characterization primarily relies on the hemagglutination inhibition (HAI) assay developed in
the 1940s. 1578 Though simple and inexpensive, HAI assays have several significant drawbacks that
compromise their utility and reliability for antigenic characterization.1579,1580 First, viruses may acquire
adaptive mutations that alter antigenicity during isolation in eggs or cells, in which case the HAI assay
will not report on the true antigenicity of the virus present in the original clinical sample. Second, HAI
assays are not standardized and exhibit significant variability in the results obtained by different
laboratories.1581,1582,1583 Finally, technical issues preclude the use of the HAI assay to characterize the
antigenicity of many recent H3N2 viruses. Alternative assays for antigenic characterization are currently
used when HAI results are difficult to interpret but are more time-consuming and technically demanding
than the HAI assay. Due to the time pressures faced by WHOCCs, particularly in the time period
immediately preceding the bi-annual VCMs, neither alternative is a viable replacement for the HAI
1574 (2015z) Interview with Centers for Disease Control and Prevention representative.
1578 Hirst GK (1942) THE QUANTITATIVE DETERMINATION OF INFLUENZA VIRUS AND ANTIBODIES BY MEANS
OF RED CELL AGGLUTINATION. J Exp Med 75: 49-64

assay.1584 Notably, even the length of time needed for shipping samples from NICs to WHOCCs (e.g. two
to three months between 2010 and 2012 in the WHOCC London region) precludes consideration of
isolates collected close to the VCM dates in strain selection decisions.1585 These exclusions effectively
lengthen the period of time between strain selection and the target flu season, which may adversely affect
vaccine match.
GoF approaches have potential to benefit the surveillance of human seasonal influenza viruses by
facilitating the prediction of antigenic phenotype directly from genotype in two ways. First, HA
sequences can be inspected for the presence or absence of molecular markers for antigenic drift that were
identified through GoF approaches. Second, that same GoF-derived data can be used to improve existing
models for predicting antigenicity based on genotype. In either case, that information could supplement
phenotypic characterization data, to strengthen the certainty of conclusions about antigenic relationships
between strains, or could be used in lieu of phenotypic characterization data. These proximal benefits of
GoF to seasonal influenza surveillance may ultimately increase the efficacy of seasonal influenza
vaccines by improving strain selection capabilities, discussed further below. In brief, GoF approaches that
enable prediction of antigenic phenotype from genotype may improve the quality of the input data used
for strain selection decisions by increasing the robustness of antigenic characterization data and, if
sequencing is performed on clinical isolates, providing information about the natural antigenicity of
strains. Additionally, because sequence data can be collected rapidly and economically and is increasingly
being generated at NICs, use of sequence-based approaches for determining antigenicity may increase the
quantity of data that can be considered, in particular from the time period immediately prior to VCM
meetings. Together, improvements to the quantity and quality of input data upon which strain selection
decisions are based will increase the likelihood that recommended strains match those that are circulating
during the target flu season, which results in increased vaccine efficacy.
During the current strain selection process, HA sequences are inspected for the presence of amino acid
substitutions that are known to be associated with altered antigenicity. Structural modeling may be used
to help predict whether the substitution will alter antigenicity in that particular genetic context.1586,1587
This information can be used to corroborate antigenic characterization data from the HAI assay or can
help to resolve antigenicity questions when HAI assay results are difficult to interpret. While this
information informs the decision-making process, the utility of these markers is limited by significant
uncertainties in the state of this science. First, the ability to reliably predict whether a particular amino
acid substitution will confer antigenic change in a new genetic context is poor. Second, because other, as-
yet-undiscovered amino acid changes may alter antigenicity, the absence of known markers is not yet
meaningful (i.e. does not indicate that the antigenicity of the strain is unchanged).
GoF approaches are critical for addressing both aspects of scientific uncertainty to strengthen the utility of
molecular marker data for antigenic change. To strengthen the predictive value of molecular markers for
antigenic change, several types of experiments are needed:
1585 Ibid.

• Targeted mutagenesis to introduce known genetic markers for altered antigenicity into new
genetic contexts (i.e. validate the antigenic consequences of the marker in a variety of strain
contexts), which represents a GoF approach,
• Targeted mutagenesis to determine which amino acid substitutions at a particular site previously
associated with antigenic change are sufficient to alter antigenicity, which represents a GoF
approach, and
• Experiments that explore the antigenic plasticity of the HA protein, to discover new substitutions
that confer antigenic change as well as substitutions that do not alter antigenicity.
To address the third experimental goal, two GoF approaches (serial passaging and forward genetic
screens) are capable of uncovering novel mutations that confer antigenic change, and targeted
mutagenesis can be used to confirm their causality (also GoF). Although these data will undoubtedly
strengthen the predictive value of molecular markers for antigenic change, given the importance of
genetic context on influenza biology, significant challenges face any effort to improve the predictive
value of such markers to a level that is meaningful. In large part, this barrier derives from the fact that the
antigenic plasticity of the HA protein is undefined. If HA can accept a very large number of amino acid
substitutions, determining the range of substitutions that do and do not alter antigenicity in a variety of
strain contexts is likely to be difficult, if not impossible. If the number of substitutions that HA can accept
is limited, then delineating this set of substitutions may be feasible. Influenza researchers felt that results
from a limited number of additional GoF experiments, to explore whether known markers are conserved
and to define the mutational landscape of antigenic drift, are likely to provide insight into the question of
whether this goal is achievable. Finally, the fact that negative results are generally not published in the
scientific literature also hinders advancements in this area, as knowing when markers are not conserved
critically informs their utility.
GoF data can also be used to improve the quality of computational models for predicting antigenic
phenotype from genotype, which represents a different sequence-based approach for predicting
antigenicity. Current models cannot accurately predict antigenic phenotype from genotype.1588 GoF
approaches have potential to improve these models in two ways: (1) by generating experimental data
about novel antigenic changes that are necessary and sufficient to alter antigenicity, which can be
incorporated into datasets used to train the models, and (2) by testing predictions of novel mutations that
would affect antigenicity that these models make, the results from which will feed back to improve model
accuracy. As existing models are primarily trained using historical data (i.e. the sequences and antigenic
characterization data from historical isolates), the ability of GoF approaches to explore new antigenic
space will complement existing data sources to enhance the predictive capability of these models for
currently circulating isolates that are evolving antigenically in new ways. As above, the feasibility of
developing models that can accurately predict antigenic phenotype from genotype will depend on the
antigenic plasticity of the HA protein and other factors, which is currently unknown.
If the landscape of amino acid substitutions that can give rise to antigenic change is large, then molecular
markers and computational models may never be robust enough to replace antigenic characterization data
generated through laboratory assays. Nonetheless, given the shortcomings of phenotypic assays for
characterizing antigenicity, the ability to corroborate laboratory results using sequence-based predictions
can significantly strengthen the quality of antigenic characterization data, particularly if clinical
specimens are directly sequenced.

As described above, GoF approaches have the potential to benefit antigenic surveillance for human
seasonal influenza viruses in two ways: (1) by improving the predictive value of molecular markers for
antigenic drift, and (2) by improving the accuracy of models for predicting antigenic phenotype from
genotype. The ability of alternative experimental approaches to similarly strengthen the utility of
molecular marker data and predictive models is evaluated to understand whether alt-GoF approaches have
the potential to benefit surveillance through either mechanism.
Currently, the predictive value of molecular markers for antigenic drift is limited by three sources of
scientific uncertainty: (1) whether markers alter antigenicity in different genetic contexts, (2) whether
novel amino acid substitutions at particular sites that are known to be associated with antigenic drift will
alter antigenicity, and (3) what other amino acid substitutions confer antigenic change. Characterizing the
antigenicity of wild type viruses that contain known molecular markers can demonstrate whether a known
marker is associated with altered antigenicity in a new genetic context, but no alt-GoF approaches are
capable of validating that the marker is necessary and sufficient to confer antigenic change in a new
strain, which is essential for application of that knowledge to surveillance.1589 Similarly, characterization
of wild type viruses is limited to determining whether different mutations at known sites or novel
mutations are associated with antigenic change. Given the limited accuracy of existing models,
predictions of any type must be experimentally confirmed using GoF approaches. Finally, as described in
section 16.5.3.1.3, GoF approaches are uniquely capable of defining the antigenic plasticity of the HA
protein, which will determine the feasibility of using molecular marker data to infer antigenic phenotype
from genotype at all. However, in all cases, attenuated reassortant strains can be used in lieu of wild type
strains because the antigenicity of the 6:2R strain is similar to that of the parental wild type strain.
Existing models for prediction of antigenic phenotype from genotype are largely built and validated using
historical data. Though comparative analysis of additional historical sequences may uncover new amino
acid substitutions that are associated with antigenic change, such data are unlikely to improve the ability
of models to predict the antigenic phenotype of currently circulating viruses, which are evolving in new
ways, and also cannot be used to validate those predictions. Thus, unlike GoF approaches, alt-GoF
approaches are unable to substantially improve existing models by generating new experimental data
about relationships between antigenic phenotype and genotype in a variety of strain contexts. However,
several completely different types of data can increase the accuracy of these models and will complement
improvements that can be gleaned through the use of GoF data. These additional data sources include
crystal structures for the HA proteins from a wider variety of strains as well as data about how various
amino acid substitutions affect HA stability, which can be generated using in vitro, virus-free
approaches.1590
GoF approaches that lead to evasion of existing natural or induced immunity have potential to benefit
surveillance of human seasonal influenza viruses in two ways: by increasing the utility of molecular
markers for antigenic drift and by improving the accuracy of existing models for predicting antigenic
phenotype from genotype. Attenuated reassortant strains (i.e. 6:2R strains with PR8) can be used in lieu
of wild type strains without diminishing these benefits.

GoF approaches are uniquely capable of discovering new amino acid substitutions that are necessary and
sufficient to alter antigenicity as well as determining whether markers are conserved in different strain
contexts, which collectively increase the predictive value of molecular markers for antigenic change.
Given the importance of genetic context for antigenic phenotype, whether such markers will ever be
strongly predictive is unknown. However, GoF approaches to explore the antigenic plasticity of the HA
protein are uniquely capable of addressing that question. Alternative experimental approaches cannot
provide causative data on molecular markers that contribute to altered antigenicity and are limited to
studying antigenic changes that have already occurred in nature, which significantly limits their utility for
this application.
GoF approaches are uniquely capable of generating experimental data about novel mutations that are
necessary and sufficient to confer antigenic change as well as validating predictions about antigenic
phenotype based on the sequences of currently circulating viruses, which will improve the accuracy of
existing predictive models that are largely based on historical data. However, alternative types of data,
including crystal structures of HA proteins from additional strains, are also needed to improve the quality
of existing models and will complement gains achieved through the use of GoF approaches.
Together, molecular markers for antigenic change or predictive models can be used to supplement or
replace lab-generated antigenic characterization data used to recommend strains for inclusion in the
seasonal influenza vaccine. The strengths and limitations of using molecular markers or predictive models
for antigenic evaluation of surveillance isolates, relative to the use of phenotypic assays, are summarized
in Table 15.29. Although molecular marker data currently informs strain selection decisions, neither data
source is robust enough to replace phenotypic data (and may never be). However, use of these data to
supplement phenotypic data has potential to improve the quantity, timeliness, and quality of antigenic
characterization data that can be considered during VCM meetings, which will ultimately increase the
likelihood that recommended strains match those that are circulating during the target flu season, thereby
leading to increased vaccine efficacy. Because molecular marker data are currently used in the strain
selection process, new data can be seamlessly incorporated into the existing process, so that the only
barrier to realization of this benefit is the need to strengthen the state of the science. Influenza researchers
involved in the strain selection process stated that computational modeling could play an important role as
well, once existing models are improved.1591 Notably, realization of all GoF benefits to antigenic
surveillance relies on the generation of sequence data directly from clinical samples and at NICs, which
enables antigenic evaluation earlier than if viruses are shipped to WHOCCs for laboratory-based
antigenic characterization. About one-quarter to one-half of HA sequences were generated at NICs during
the 2014 – 2015 flu season (discussed in detail below), and sequencing of clinical samples is carried out
and is increasingly common (see section 15.3.4.2.1), demonstrating that these GoF benefits can be
realized immediately. 1592 However, full realization of these benefits necessitates expanding sequencing
capabilities at NICs and increasing the number of clinical samples that are directly sequenced.
1592 (2015n) Personal communication from WHOCC representative.

Table 15.29. Summary of the Benefits of GoF Approaches that Lead to Evasion of Existing Natural or Induced Immunity
Surveillance benefits – Aid evaluation of the antigenicity of circulating seasonal influenza viruses
• Could increase the quality of phenotypic characterization data:

o Clinical samples can be directly sequenced
o Corroboration of HAI assay results increases the robustness of the
data • Scientific uncertainties compromise the
current utility of molecular markers for
• Could increase the quantity of antigenic characterization data
antigenic change
GoF #1: considered during VCM meetings:
o Time frame for establishing that knowledge
o As sequencing becomes cheaper and easier, a greater number of
Strengthen the predictive is uncertain, likely to be long-term
viruses may be sequenced than subjected to lab-based antigenic
value of molecular markers • Use of molecular markers is inherently
characterization
for antigenic change predictive
o Because NICs are increasingly capable of sequencing virus samples,
sequence-based evaluation enables consideration of isolates collected • Full realization of benefits depends on
immediately prior to VCM meetings expanding sequencing capabilities at NICs
• Molecular marker data are currently used to interpret seasonal flu
surveillance data
o New data can be incorporated into the process in the immediate term
• Could increase the quality of phenotypic characterization data:
o Clinical samples can be directly sequenced • Reliable computational models for predicting
o Corroboration of HAI assay results increases the robustness of the antigenicity based on sequence do not yet
data exist, and their future utility depends on
GoF #2:
scientific advancements
Support development of • Could increase the quantity of antigenic characterization data
o Timeframe for establishing that knowledge
computational models for considered during VCM meetings:
is uncertain, likely to be long-term
predicting antigenicity o As sequencing becomes cheaper and easier, a greater number of
viruses may be sequenced than subjected to lab-based antigenic • Use of computational models is inherently
based on sequence predictive
characterization
o Because NICs are increasingly capable of sequencing virus samples, • Full realization of benefits depends on
sequence-based evaluation enables consideration of isolates collected expanding sequencing capabilities at NICs
immediately prior to VCM meetings

Surveillance benefits – Aid evaluation of the antigenicity of circulating seasonal influenza viruses
• Viruses may acquire adaptive mutations that
alter antigenicity during isolation in eggs or
cells, rendering results inaccurate
• HAI assays are not standardized and exhibit
significant variability in the results obtained
by different laboratories
Alt-GoF #1:
• Technical issues preclude the use of the HAI
assay to characterize the antigenicity of many
Phenotypic evaluation of recent H3N2 viruses
antigenicity using the HAI
• Provides a direct readout of the antigenicity of a given virus
• Alternative lab assays for antigenic
assay or other assays
characterization are time-consuming and more
technically demanding
• The time needed to ship samples from NICs to
WHOCCs for antigenic characterization
delays generation of the data
o Many isolates are not shipped in time for
consideration at VCM meetings

15.5.5 GoF Benefits to the production of vaccines
15.5.5.1 Vaccine development benefit #1: Improve strain selection capabilities for seasonal influenza
vaccines
Antigenic drift of human seasonal influenza viruses necessitates frequent updating of influenza vaccines.
Since the early 1970s, the WHO has provided formal recommendations for the strain composition of
seasonal influenza vaccines based on year-round influenza surveillance conducted through the GISRS
(described above).1593,1594 Based on analysis of the genetic, antigenic, and epidemiologic characteristics of
several thousand influenza isolates collected throughout the year, experts suggest candidate vaccine
viruses that are likely to antigenically match the strains that will be circulating during the target flu
season.1595,1596,1597 Because of the long production timescales for influenza vaccines (six to eight months),
recommendations must be made nearly one year in advance of the predicted peak of influenza activity for
the target season.1598,1599 Despite the complexity of the data considered and the challenge of predicting
dominant strains many months in advance, this process generally works well. Most years, the vaccine is
well-matched to circulating strains, capable of preventing influenza-like-illness in approximately 70% of
vaccine recipients aged 15 – 64 years.1600 However, occasionally a rare antigenic variant rises to
prominence during the course of vaccine production, as happened during the recent 2014 – 2015 flu
season for the H3N2 strain, which results in poor vaccine match and reduced vaccine efficacy.1601,1602
Several shortcomings compromise the efficacy of the current strain selection process. First, the timeliness
and representativeness of isolates forwarded to WHOCCs by NICs, which form the basis of strain
selection recommendations, could be improved. Due to significant lag times between sample collection
and shipment (e.g. two to three months between 2010 and 2012 in the WHOCC London region), many
isolates cannot be analyzed in time for consideration during VCM meetings, which effectively lengthens
the period of time between strain selection and the target flu season. Additionally, the viruses that are
forwarded may not be fully representative in terms of geography, climate, age groups, and epidemic
timing, due to reductions in the number of hospitals that submit samples to NICs and other funding
challenges. Taken together, these shortcomings in existing surveillance networks reduce the quality and
1593 Oshitani H (2010) Influenza surveillance and control in the Western Pacific Region. Western Pacific surveillance and
response journal : WPSAR 1: 3-4
1594 WHO. Process of Influenza Vaccine Virus Selection and Development
http://apps.who.int/gb/pip/pdf_files/Fluvaccvirusselection.pdf. Last Update November 19, 2007. Accessed November 22,
2015.
1595 Trivalent influenza vaccines (most common) include one A/H1N1, one A/H3N2, and one B strain. Quadrivalent influenza
vaccines include an additional B strain.
1597 Stöhr K (2013b) Influenza vaccine production. In Textbook of Influenza, Frs RGW, Md ASM, Md TJB, ScD RAL (eds), pp
352-370. John Wiley & Sons, Ltd
1600 Legrand J et al (2006) Real-time monitoring of the influenza vaccine field effectiveness. Ibid. 24: 6605-6611
1602 Xie H et al (2015) H3N2 Mismatch of 2014-15 Northern Hemisphere Influenza Vaccines and Head-to-head Comparison
between Human and Ferret Antisera derived Antigenic Maps. Sci Rep 5: 15279

quantity of input data for strain selection decisions, which compromises the accuracy of the process. A
second shortcoming of the current strain selection process is its heavy reliance on the HAI assay for
antigenic characterization of surveillance isolates, which suffers several significant drawbacks as detailed
above in section 15.5.4.1. A final shortcoming is the inability to reliably predict whether rare antigenic
variants will rise to prominence in nature during the vaccine production process, which results in poor
vaccine match.
GoF approaches that lead to evasion of existing natural or induced immunity have potential to address all
three shortcomings in the current strain selection process.
First, as discussed in section 15.5.4.1, GoF approaches have potential to strengthen the predictive value of
molecular markers for antigenic drift and to improve the accuracy of existing models for predicting
antigenic phenotype from genotype. Either strategy for sequence-based prediction of antigenic phenotype
could be used to corroborate lab-generated HAI data in cases where results are difficult to interpret. This
supplemental data source could strengthen the robustness of antigenic characterization information,
thereby improving the quality of input data for the strain selection decision. Alternatively, sequence-based
prediction methods could replace laboratory methods for antigenic characterization. Given that sequence
data can be collected rapidly and economically and is increasingly being generated at NIC labs, reliance
on sequence data may allow for consideration of a greater number of isolates, including isolates sampled
close to the VCM dates. The result, an increase in the quantity of input data for the strain selection
decision, would improve the process through a different mechanism. Critically, although molecular
marker data informs strain selection decisions, neither molecular marker data nor predictive models are
currently robust enough to replace phenotypic data (and may never be). Notably, GoF approaches are
uniquely critical for advancing the state of the science for both approaches, although other types of data
are also needed to improve predictive models and will complement GoF data on molecular markers of
antigenic drift. During the 2014 – 2015 season, an average of 28 – 44% of HA sequences were generated
at NICs, depending on the strain (range 0% to 70%), though only 9 – 13% of those sequences were
submitted in time for consideration in the February VCM meeting.1603 Thus, given current diagnostic
capabilities at NICs, GoF benefits to sequence-based prediction of antigenicity can be realized in the
context of the current surveillance system. However, full realization of this benefit necessitates the
expansion of sequencing capabilities at NICs as well as increasing the timeliness of sequencing data
generated at NICs.
Second, GoF approaches to experimentally induce drift can be used to predict how circulating viruses
may drift in nature, enabling production of vaccines against future, “drifted” strains that will antigenically
match circulating viruses at their time of deployment. Specifically, the selection of antibody escape
mutants of currently circulating viruses, through serial passaging or forward genetic screens conducted in
vitro and in vivo, enables the identification of HA substitutions that confer escape. Coupled with genetic
surveillance data, this information can be used to forecast the antigenicity of the next dominant strain to
arise in nature.1604,1605 However, whether and when such variants will emerge is uncertain, in part because
stochastic events in natural evolution may result in the appearance of an unusual mutant that was not
selected in the experimental studies. For that reason, this data is not currently incorporated into the strain
selection process, and additional research is needed to determine whether it will be useful for predicting
1603 (2015n) Personal communication from WHOCC representative.


the course and timing of antigenic evolution in the future.1606 Additionally, researchers emphasized that if
this strategy is implemented in the strain selection process, the knowledge base must be regularly updated
by performing experiments with currently circulating strains.
Finally, a different approach for predicting antigenic drift involves the use of computational models for
antigenic evolution (though computational models could be used in conjunction with experimental data).
Existing models for prediction of antigenic drift are built largely using historical data (including paired
sequence and antigenic data generated for the purpose of strain selection) and have been validated using
historical data.1607,1608 As a result, their prospective applicability and utility are unknown. Two types of
GoF studies are needed to improve the quality of existing models. First, a better understanding of the
process of antigenic evolution will provide a foundation for the design of better models. As described
above (section 15.5.3.2), GoF approaches are uniquely capable of providing in-depth information about
the evolutionary mechanisms driving antigenic drift as well as prospective information about the
evolution of currently circulating viruses, although results may not translate to antigenic evolution of
viruses in human populations in nature. Second, influenza modeling experts have stated that developing
the ability to predict whether particular amino acid substitutions alter antigenicity in a given genetic
context is critical for advancing the quality of these models.1609,1610 As described in the preceding section,
GoF approaches are essential for improving the accuracy of models for prediction of antigenic phenotype
from genotype, although other types of data are also needed.
Taken together, utilizing experimental and/or in silico approaches to predict whether new antigenic
variants are likely to emerge during the course of vaccine production would enable the production of
vaccines based on those predicted future strains. This strategy would increase the likelihood that vaccines
match the strains that are circulating during their target flu season, which will lead to an overall
improvement in vaccine efficacy. One key concern associated with this strategy is that evolutionary
predictions are difficult and are unlikely to be correct one hundred percent of the time, even as the science
of prediction advances. Importantly, the exact amino acid sequence of the next dominant strain does not
need to be predicted, but rather its antigenicity (as multiple sequences can fall into the same antigenic
“cluster”). In addition, studies have shown that immunization with “antigenically advanced” vaccines, i.e.
those that are based on predicted future strains, can protect against currently circulating strains. That is, in
addition to stimulating production of new antibodies against the antigenically advanced vaccine strain,
vaccination re-stimulates production of old antibodies produced in response to prior vaccines, an effect
termed “immunity back-boost.”1611 Thus, even if the prediction is incorrect (i.e. the strain does not drift in
nature), pre-emptive vaccination strategies are likely afford some degree of protection.
15.5.5.1.2 Benefits and limitations of alt-GoF approaches for improving strain selection capabilities
As described above, comparative sequence analysis and in silico approaches are capable of identifying
new molecular markers that are associated with antigenic change or are predicted to alter antigenicity,
respectively. However, that such markers are necessary and sufficient to cause antigenic change across a
variety of influenza strains must be confirmed through GoF experiments for these data to be applied to the
1607 Bedford T et al (2014) Integrating influenza antigenic dynamics with molecular evolution. Elife 3: e01914
1608 Du X et al (2012) Mapping of H3N2 influenza antigenic evolution in China reveals a strategy for vaccine strain
recommendation. Nat Commun 3: 709
1611 Fonville JM et al (2014) Antibody landscapes after influenza virus infection or vaccination. Science 346: 996-1000

interpretation of genetic surveillance data. Furthermore, alt-GoF approaches are constrained to studying
antigenic changes in nature. For these reasons, neither alternative approach provides data that can
strengthen the predictive value of molecular markers for antigenic change or improve existing models for
predicting antigenic phenotype from genotype.
Comparative sequence analysis can also provide insight into antigenic evolution, which critically
complements laboratory evolution studies by generating insights that are directly relevant to the evolution
of flu viruses in human populations in nature. However, the ability of comparative sequence analysis to
provide mechanistic information about evolution is severely limited relative to GoF approaches. In
addition, this alt-GoF approach cannot provide prospective information about the evolution of currently
circulating viruses, which is the purpose of using antigenic evolution models to inform the strain selection
process. For both reasons, the use of comparative sequence analysis approaches is not sufficient to
improve the quality of existing models for antigenic evolution.
In addition to using sequence-based prediction of antigenic phenotype to complement or replace the

traditional HAI assay, alternative strategies for improving the antigenic characterization data for
surveillance isolates have been pursued. The Consortium for the Standardization of Influenza
Seroepidemiology (CONSISE), aims to standardize seroepidemiology for influenza and other respiratory
pathogens by developing and publishing consensus laboratory assay protocols, including a protocol for
the HAI assay.1612 Ultimately, these efforts have potential to improve the quality of antigenic data
considered during strain selection decisions by ensuring that antigenic data generated at disparate sites are
more comparable. A second effort to improve antigenic characterization data involves the development of
alternative antigenic characterization assays based on synthetic glycan-coated beads or solid matrices in
lieu of the red blood cells that are used for traditional HAI assays. Though these assays have greater
potential for standardization and automation than the HAI assay, alternative assays to date have had
limited success. Because of the acute time pressure faced by WHOCCs, particularly leading up to VCMs,
replacement of the HAI assay with the more time-intensive but also more accurate virus neutralization or
micro-neutralization assays is not practical.1613
Several alternative approaches have potential to improve the strain selection process through completely
different mechanisms. First, increasing the timeliness, representativeness, and availability of surveillance
isolates would improve the accuracy of strain selection decisions by augmenting the quality of the input
data upon which those decisions are based. Key elements of efforts to strengthen influenza surveillance
systems include improving national surveillance systems, public health laboratories, and reporting and
virus sharing procedures in developing countries.1614 To that end, between 2004 and 2014, the CDC
invested more than $150 million toward building sustainable lab capacity and NICs and other
international laboratories in over 40 less developed countries around the world, such as India, Cambodia,
Vietnam, and Egypt. The CDC also works closely with Ministries of Health to ensure that they are
conducting epidemiological surveillance, including the collection of “metadata” about patient
demographics, whether patients have been treated with antivirals or were vaccinated, and other factors
along with clinical samples.1615 The WHO and other WHO member countries also provide support, in the
form of funding, technical expertise, and guidance. However, given that resources for public health are
1612 Van Kerkhove MD et al (2013) The consortium for the standardization of influenza seroepidemiology (CONSISE): a global
partnership to standardize influenza seroepidemiology and develop influenza investigation protocols to inform public health
policy. Influenza Other Respir Viruses 7: 231-234
1614 Ibid.
1615 (2015z) Interview with Centers for Disease Control and Prevention representative.

limited and governments have many competing priorities, sustaining and building upon gains in these
areas that have occurred in the wake of the 2009 pandemic will continue to pose a major challenge.1616,1617
Other lines of research and new technologies have potential to fundamentally change current influenza
virological surveillance strategies and activities and may also lead to improved strain selection. For
example, an improved understanding of the spatiotemporal distribution of viruses and the factors that
influence the geographic spread of viruses could help target surveillance efforts and may also inform
prediction of whether and when antigenic variants detected in a particular region are likely to arise.1618
Deep sequencing of surveillance isolates and systems biology approaches to analysis of such data may
provide insight into the role of host-pathogen interactions in the antigenic evolution of viruses, which
could also influence vaccination strategies and the strain selection process.1619 In these and other cases,
because the state of the science and/or technology is preliminary, whether and when these approaches will
have a demonstrated impact on strain selection for seasonal influenza vaccines is unknown.
15.5.5.1.3 Benefits and limitations of alternative approaches for improving the efficacy of seasonal flu
vaccines through other mechanisms
In addition to improving strain selection capabilities, several completely different strategies can be used
to increase the efficacy of seasonal flu vaccines. These strategies are described in detail in section
15.2.4.3.4 and are briefly summarized here. First, a universal or broad-spectrum flu vaccine would
obviate the need for yearly production of strain-specific vaccines. However, influenza and vaccinology
experts disagree about the scientific feasibility of developing a universal vaccine, and one expert felt that
a ten to twenty year time frame for development is optimistic. Second, several scientific and technical
advancements could shorten production timelines for strain-specific vaccines, which would enable strain
selection closer to the start of flu season, presumably increasing the likelihood that the correct strains will
be chosen. New vaccine platforms, such as recombinant vaccines, can be rapidly scaled up and have
shorter production timelines than egg- and cell-based vaccines. However, the one recombinant vaccine on
the market accounts for less than 1% of total seasonal influenza vaccine produced annually, and although
several other virus-free vaccine platforms are in development, the length and expense of licensure
processes for new vaccines will delay their widespread availability. In addition, it is unclear at what point
virus-free vaccines will make up a large enough market share that strain selection meetings could be
shifted back (which would compromise the ability of egg- and cell-based vaccine manufacturers to
produce vaccine in time for the start of flu season). Incorporating adjuvants into existing egg- and cell-
production of the same number of doses in a shorter period of time. However, no US-licensed seasonal
vaccines include adjuvants. Although an active area of research, adjuvanted vaccines must undergo
standard FDA licensing procedures for new vaccines and thus are unlikely to be broadly available in the
near future. Finally, GoF research that enhances virus production enables the development of higher-yield
CVVs, which shortens vaccine production timelines by increasing the rate of bulk antigen production. It
should be noted that manufacturers already initiate production of at least one component of the seasonal
vaccine “at risk” in advance of the VCM meeting, in order to produce sufficient vaccine by the start of flu
season. For that reason, it is not clear whether the ability to shorten production timelines for egg- and cell-
based vaccines would trigger a shift in the timing of the VCM or would lead manufacturers to delay
1618 Ibid.
1619 Ibid.

initiation of bulk antigen production so that all components are produced after the meeting results are
publicized.
Taken together, GoF approaches are uniquely capable of strengthening the predictive value of molecular
markers for antigenic change and play a critical role in improving models for predicting antigenic
phenotype from genotype as well as models for predicting antigenic drift. Although alternative
experimental approaches can provide other types of data that also strengthen predictive models, these data
complement rather than replace GoF data.
Advancing capabilities in these areas has the potential to benefit the strain selection process for seasonal
influenza vaccines in several ways, summarized in Table 15.30. First, using sequence-based prediction of
antigenic phenotype to reinforce HAI assay results strengthens the robustness of antigenic
characterization data, which provides a foundation for strain selection decisions. Second, given that
genetic surveillance data are increasingly available from NICs and other sample collection sites, shifting
to sequence-based prediction of antigenic phenotype in lieu of laboratory assays has potential to increase
the timeliness and quantity of surveillance data that are considered during VCMs. Third, predicting
antigenic drift using models or through experimental GoF approaches would enable the development of
antigenically advanced vaccines that are likely to match the circulating strains when vaccines are
deployed, thereby increasing vaccine efficacy.
Current experimental and modeling efforts cannot yet predict antigenic phenotype from genotype or the
timing and direction of antigenic drift. Whether and when such capabilities will be sufficiently accurate to
be incorporated into the strain selection process is unknown and depends both on scientific advancements
and inherent features of influenza biology. Namely, the antigenic plasticity of the HA protein is not well-
characterized but governs the feasibility of each of these predictive efforts. Notably, GoF efforts are also
essential for advancing understanding of the antigenic landscape of HA.
Several alternative approaches have potential to improve the strain selection process through different
mechanisms. First, efforts to standardize the HAI assay and to develop variant antigenic characterization
assays based on synthetic glycans are ongoing, in order to improve the quality of antigenic
characterization data upon which strain selection decisions are based. However, these alternative assays
are not yet viable replacements for the HAI assay, and the degree to which increased standardization of
the HAI assay will improve data quality is uncertain. Initiatives to strengthen global influenza
surveillance systems have potential to improve the timeliness, representativeness, and quantity of
surveillance isolates that can be considered at VCM meetings but face considerable funding and political
barriers. Finally, new technologies such as deep sequencing have the potential to revolutionize influenza
virological surveillance activities and may improve strain selection capabilities through unexpected
mechanisms. Each of these alternative approaches either complements GoF approaches or addresses
different shortcomings in the strain selection process.
Given the complexities involved in coordinating global influenza surveillance and making strain selection
decisions under the time pressures imposed by vaccine production timelines, as well as the significant
uncertainties in whether and when both GoF and alt-GoF approaches will yield demonstrable benefits to
the process, pursuing both GoF and alt-GoF strategies in tandem will ensure that strain selection
capabilities are advanced rapidly and to the greatest extent possible.
Finally, several alternative approaches have potential to improve the efficacy of seasonal influenza
vaccines through completely different mechanisms. The strengths and limitations of these approaches
relative to strategies for improving strain selection capabilities are summarized in Table 15.13. Universal

vaccines represent the only strategy with potential to fully “solve” the vaccine mismatch problem but are
in early stages of development and represent a long-term solution at best. Several approaches, namely the
development of virus-free vaccines, the incorporation of adjuvants into existing vaccines, and the
development of higher-yield vaccine viruses through GoF approaches that enhance virus production, have
potential to shorten production timelines for strain-specific vaccines. This adjustment to manufacturing
schedules could enable strain selection closer to the start of flu season, which presumably will increase
the likelihood of vaccine match. Importantly, all of these approaches complement efforts to improve
strain selection capabilities because each approach addresses different underlying gaps in current
scientific and technical capabilities that contribute to vaccine mismatch. Thus, influenza vaccine experts
recommend pursuing all of these approaches as part of comprehensive strategy for improving the quality
of seasonal influenza vaccines.

Vaccine benefits – Increase the efficacy of seasonal flu vaccines by improving strain selection capabilities
• Improves the quality of antigenic characterization data upon which strain • Scientific uncertainties compromise the
selection decisions are based: current utility of molecular markers for
GoF #1: o Clinical samples can be directly sequenced antigenic change, and existing models for
Improve methods for o Corroboration of HAI assay results increases the robustness of the prediction of antigenicity are not accurate
predicting antigenicity based data o The time frame for advancing the state
on sequence data: • Increases the quantity of antigenic characterization data upon which of science in both areas is uncertain, and
is likely to be long-term
• Strengthen the predictive strain selection decisions are based:
value of molecular o As sequencing becomes cheaper and easier, a greater number of • Use of molecular markers or computational
markers for antigenic viruses may be sequenced than subjected to lab-based antigenic models is inherently predictive
change characterization o Unlikely to ever replace phenotypic
• Improve computational o As the number of NICs that generate sequence data increases, data, which limits ability of this
models for sequence- sequence-based evaluation enables consideration of isolates collected approach to increase the quantity of
based predictions of close to VCM dates antigenic characterization data
antigenicity considered
• Molecular marker data are currently used to interpret seasonal flu
surveillance data • Full realization of benefits depends on
o New data can be incorporated into the process in the immediate term expanding sequencing capabilities at NICs
GoF #2: • The prospective accuracy of experimental

Improve methods for • Enables production of vaccines based on future, drifted strains, which methods and computational models for
predicting antigenic drift • will antigenically match circulating viruses at their time of deployment prediction of drift is uncertain
• Experimentally induce • Immunization with “antigenically advanced” vaccines can protect against o The time frame for advancing the state
drift in circulating viruses currently circulating strains of the science in both areas is uncertain,
and may be long-term
• Improve computational o Will achieve some degree of protection even if predictions are
models for predicting incorrect • Neither approach is currently incorporated
antigenic drift into the strain selection process

Vaccine benefits – Increase the efficacy of seasonal flu vaccines by improving strain selection capabilities
Alt-GoF #1: • Improve the quality of antigenic characterization data upon which strain
Improve laboratory- selection decisions are based
generated antigenic  Standardization of HAI assay or development of alternative,
• Efforts to develop new antigenic
characterization data: characterization assays have had limited
standardized assays
success to date
• Standardize HAI assay • Improve the quantity of antigenic characterization data upon which strain • Standardization of the HAI assay is
• Develop alternative assays selection decisions are based:
challenging
for antigenic o Development of alternative assays that are higher-throughput than the
characterization HAI assay
• Increase the timeliness, representativeness, and availability of • Resources for public health are limited and
Alt-GoF #2: surveillance isolates, which will increase the quality of the antigenic governments have many competing
Strengthen global influenza characterization data upon which strain selection decisions are based priorities
surveillance networks o Improve national surveillance systems, public health laboratories, and o Maintaining and expanding current
reporting and virus sharing procedures in developing countries surveillance capabilities is challenging
Alt-GoF #3: • May help target surveillance efforts, thereby increasing the quality of
Alternative lines of research: antigenic characterization data upon which strain selection decisions are
based • The state of the science in these areas is
• Improved understanding
of the spatiotemporal • May inform predictions of whether and when antigenic variants detected preliminary
distribution of viruses and in particular regions are likely to arise, thereby enabling development of o Whether, to what extent, and when these
factors that influence better-matched vaccines approaches will benefit the strain
geographic spread • Provide insight into the role of host-pathogen interactions in the selection process is unknown
• Deep sequencing of antigenic evolution of viruses, which could influence strain selection
surveillance isolates decisions

15.5.5.2 Vaccine development benefit 2: Inform development of universal or broad-spectrum flu
vaccines
Because existing influenza vaccines are strain-specific, there is a continued need for production of new
influenza vaccines to protect public health. Specifically, seasonal influenza vaccines must be updated
annually to accommodate antigenic drift of circulating influenza viruses, and specific vaccines must be
produced in response to the emergence of a novel pandemic strain. The long production timescales for
current influenza vaccines compromise the quality of seasonal flu vaccines (vis-à-vis the potential for
reduced vaccine match) and the availability of flu vaccines during a pandemic (discussed in detail in
section 15.2.4.1). For those reasons, researchers are actively pursuing the development of broad-spectrum
flu vaccines, which could protect against multiple strains (a subset of related strains within a subtype, an
entire subtype, or multiple subtypes), and “universal” flu vaccines, which could protect against all strains.
Either type of vaccine would eliminate the need for an exact match between vaccine strains and
circulating seasonal viruses, thus improving the efficacy of seasonal flu vaccines. In addition, universal or
broad-spectrum vaccines could be available rapidly during a pandemic or could be used to pre-vaccinate
the population against emerging influenza strains, thereby increasing vaccine coverage during a
pandemic. However, given the high mutation rate of influenza viruses1620 and the high immunogenicity of
strain-specific regions of the HA protein,1621,1622 development of a broad-spectrum or universal vaccine
represents an extremely challenging prospect.1623,1624 Scientists are exploring multiple strategies for
development of such next-generation influenza vaccines, and both GoF and alt-GoF approaches have
potential to inform this process.
GoF approaches that aim to map the antigenic landscape of the HA protein have potential to inform the
development of broad-spectrum and universal influenza vaccines. Specifically, comprehensive forward
genetic screens to identify which substitutions the HA protein can tolerate and which of those
substitutions alter antigenicity will define the regions of the HA protein could drift (i.e. without
significantly compromising the stability of HA and the viability of the virus) as well as how those regions
can change antigenically. Defining all possible antigenic configurations of the HA protein provides a
foundation for developing a broad-spectrum vaccine (or vaccine cocktail) that protects against a large
fraction of the possible antigenic space, thus pre-empting antigenic drift in nature and eliminating the
need for annual production of seasonal flu vaccines.1625 Alternatively, defining those regions of the HA
protein that do not mutate may provide a foundation for the development of a “drift-resistant” universal
vaccine that targets those regions. Currently, whether either strategy will lead to the development of an
effective influenza vaccine is unknown. Given the possibility for compensatory mutations to overcome
fitness defects arising from antigenic escape mutations as well as the possibility for multiple mutations to
1620 Parvin JD et al (1986a) Measurement of the mutation rates of animal viruses: influenza A virus and poliovirus type 1.
1621 Gerhard W et al (1981) Antigenic structure of influenza virus haemagglutinin defined by hybridoma antibodies. Nature 290:
713-717
427
vaccines 7: 10-11
1624 (2015x) Interviews with Federal Government representative and Influenza researchers with expertise in vaccine
development.
1625 (2015w) Interview with Biomedical Advanced Research and Development Authority representative.

contribute to antigenic change,1626 comprehensive mapping of the antigenic landscape of HA may
necessitate evaluation of mutations singly and in combination, either of which represents a labor-intensive
project. Additionally, whether findings will be specific to an influenza subtype (or subset of strains within
that subtype) or will translate to other influenza subtypes is unknown.
Alternative approaches can also provide insight into which regions of HA mutate to alter antigenicity and
the spectrum of antigenic configurations the HA protein can assume. First, attenuated reassortant strains
(i.e. 6:2R strains with lab-adapted strains such as PR8, comprising the HA and NA genes from a seasonal
strain of interest and the remaining six genes from PR8) can be used for forward genetic screens in lieu of
wild type strains. As the antigenicity of 6:2R strains is preserved relative to that of the parental seasonal
flu strain, these strains are suitable for defining the landscape of antigenic configurations that are possible
for the HA protein. However, given epistatic effects, the suite of mutations that HA can “tolerate” may be
different in the context of a 6:2R strain versus the wild type strain.
Alternative experimental approaches can also be used to study the antigenic landscape of the HA protein.
Comparative analysis of historical isolates can provide insight into mutations that are associated with
antigenic drift over time. However, this approach is constrained to studying the fraction of antigenic space
that the HA protein has explored in nature. Moreover, this approach cannot provide information about
why certain substitutions have not been observed in nature (i.e. mechanisms driving negative selection),
which is important context for mapping the spectrum of substitutions that are possible. In addition, the
causative effects of mutations identified through comparative sequence analysis must be verified through
a GoF experiment. Modeling approaches can, in principle, fully explore antigenic space but cannot yet
accurately predict antigenic phenotype from genotype nor the effects of HA mutations on protein stability
or viral fitness.
Completely different types of scientific data, generated through alt-GoF approaches, can also inform the
development of universal and broad-spectrum influenza vaccines. For example, one method for
identifying conserved epitopes involves identifying broadly neutralizing antibodies by characterizing the
ability of different monoclonal antibodies to neutralize a variety of strains, followed by antibody epitope
mapping.1627 This knowledge can inform the development of multiple vaccine types. Another method
involves prediction of conserved immunogenic regions using in silico approaches, which has been used as
a basis for the development of peptide-based vaccines.1628,1629,1630 Some of these vaccine candidates
have been shown to be immunogenic in animal studies and Phase I clinical trials.1631,1632,1633 As all
universal vaccines are in early stages of development, whether these approaches will prove to be more or
less successful than GoF approaches in stimulating development of a safe, effective, and broad-spectrum
influenza vaccine is unknown.
1626 Myers JL et al (2013) Compensatory hemagglutinin mutations alter antigenic properties of influenza viruses. Journal of
virology 87: 11168-11172
1627 Zhu X et al (2013b) A unique and conserved neutralization epitope in H5N1 influenza viruses identified by an antibody
against the A/Goose/Guangdong/1/96 hemagglutinin. J Virol 87: 12619-12635
1628 Gottlieb T, Ben-Yedidia T (2014) Epitope-based approaches to a universal influenza vaccine. Journal of autoimmunity 54:
15-20
1629 Stoloff GA, Caparros-Wanderley W (2007) Synthetic multi-epitope peptides identified in silico induce protective immunity
against multiple influenza serotypes. European journal of immunology 37: 2441-2449
1630 Adar Y et al (2009) A universal epitope-based influenza vaccine and its efficacy against H5N1. Vaccine 27: 2099-2107
1631 Ibid.
1632 Pleguezuelos O et al (2012) Synthetic Influenza vaccine (FLU-v) stimulates cell mediated immunity in a double-blind,
randomised, placebo-controlled Phase I trial. Ibid. 30: 4655-4660
1633 Pleguezuelos O et al (2015) A Synthetic Influenza Virus Vaccine Induces a Cellular Immune Response That Correlates with
Reduction in Symptomatology and Virus Shedding in a Randomized Phase Ib Live-Virus Challenge in Humans. Clinical
and vaccine immunology : CVI 22: 828-835

The strengths and limitations of GoF and alt-GoF approaches that inform the development of universal or
broad-spectrum influenza vaccines are summarized in Table 15.31. GoF approaches are uniquely capable
of defining the antigenic landscape of the influenza HA protein – that is, the spectrum of antigenic
configurations that HA can assume and which regions of HA are capable of mutating while preserving
virus viability. These data may inform the development of broad-spectrum influenza vaccines, which
protect against a large fraction of the possible antigenic space, or universal influenza vaccines, which
target regions of the protein that are unable to mutate and thus are drift-resistant. Alternative experimental
approaches have significant limitations. Attenuated reassortant strains can be used to explore possible
antigenic configurations, but results regarding the fitness consequences of mutations may not translate to
wild type strains. Comparative analysis of historical isolates is limited to the fraction of antigenic space
that has been explored in nature and cannot provide information on mutations that compromise virus
viability. While virus-free approaches can be used to explore new antigenic space, these approaches do
not reveal the fitness consequences of mutations either. Finally, existing models cannot accurately predict
antigenic phenotype from genotype or predict the fitness consequences of particular mutations.
Mapping the antigenic landscape of the HA protein represents a labor-intensive project, and whether
vaccine development strategies based on the information gleaned from this approach will be successful is
unknown. Other strategies for developing broad-spectrum and universal vaccines, such as in silico
prediction of conserved epitopes for the development of peptide-based vaccines, have shown promise. All
universal/broad-spectrum vaccine candidates are in early stages of development, and which strategy is
likely to be most successful is unknown. Given the challenges for developing universal/broad-spectrum
vaccines, pursuing all experimental approaches that support vaccine development in tandem, including
GoF approaches, will maximize the likelihood of success, which could have large public health impacts.

Benefits to vaccine development: Inform the development of universal influenza vaccines
• Defining all possible antigenic configurations • Whether either strategy will enable the development
GoF #1:
of HA could enable the development of broad- of more effective influenza vaccines is unknown
Comprehensive forward genetic screens to
map the antigenic landscape of the HA
spectrum vaccines that protect against a large • Benefits are likely to be long-term
fraction of the possible antigenic space
protein o Approach is scientifically challenging and labor-
• Defining regions of the HA protein that cannot intensive
• Identify regions of HA that can drift drift could enable the development of “drift-
without compromising virus viability o Whether results will be strain- or sub-type specific
resistant” vaccines targeting those regions is unknown
Alt-GoF #1: • Attenuated reassortant strains cannot provide reliable

information about whether and to what extent
Alternative Experimental Approaches for
antigenicity-altering mutations compromise the
mapping the antigenic landscape of the HA • Attenuated reassortant strains and
viability/fitness of wild type viruses
protein computational models can be used to fully
explore antigenic space • Predictions derived from computational models must
• Use of attenuated reassortant strains be experimentally validated
• Comparative sequence analysis can provide
• Comparative analysis of historical information about substitutions that are • Comparative sequence analysis is constrained to
sequences
associated with antigenic drift over time studying the fraction of antigenic space that nature
• Computational models for prediction of has already explored
antigenic phenotype from genotype
• Cannot reveal negatively selected mutations
Alt-GoF #2:
Alternative strategies for developing
universal flu vaccines:
• Experimentally identifying broadly • Approach has generated several promising
vaccine candidates • Whether these approaches will lead to the generation
neutralizing antibodies of safe and effective vaccines is unknown
• Prediction of conserved immunogenic
regions
• Other approaches

15.6 Influenza viruses: Detailed Analysis of the Benefits of GoF Research that Leads to
Evasion of Vaccines
lead to evasion of vaccines in development. In this section, an overview of GoF approaches in this
phenotypic category is provided, and the scientific outcomes and/or products of each approach are
described.
Serial passaging of a virus in cells in the presence of animal sera produced in response to a vaccine or in
vaccinated animals may lead to the emergence of viruses that are resistant to neutralization by vaccine-
induced antibodies. This approach is used to test whether and how readily viruses can evolve to evade
vaccines in development, for example new vaccine platforms that are more broad-spectrum or resistant to
drift than current influenza vaccine platforms, which is an important indicator of the potential field
efficacy of the vaccine.
Influenza Viruses that are Resistant to Therapeutics
This GoF approach is solely focused on understanding how a virus evolves in response to immune
pressure from a vaccine under development. As a result, insights gleaned from this approach do not
benefit scientific knowledge, surveillance or policy decisions (because the vaccine has not yet been
deployed) or the development of therapeutics and diagnostics.
15.6.2.1 Vaccines
GoF approaches that lead to evasion of vaccines in development benefit the development of new
influenza vaccines. Specifically, these approaches demonstrate whether and how readily viruses can drift
to escape neutralization by new vaccine candidates, which is an important indicator of their potential field
efficacy relative to existing vaccines.
GoF benefits to the development of new vaccines may have downstream economic benefits. Economic

15.6.3 Benefits to Vaccine Development
15.6.3.1.1 Shortcomings in existing influenza vaccines
Because existing influenza vaccines are strain-specific, new seasonal flu vaccines must be produced
annually in order to accommodate antigenic drift of circulating influenza viruses, and new pandemic flu
vaccines must be produced in response to the emergence of a novel pandemic strain. The production
timeline for egg- and cell-based influenza vaccines, which comprise over 99% of seasonal flu vaccine
doses produced annually, currently spans six to nine months.1634 As a result, vaccines are unavailable until
many months into a pandemic, and the strains for the seasonal flu vaccine must be chosen six months in
advance of the start of the target flu season, which occasionally leads to vaccine mismatch and reduced
vaccine efficacy. For these reasons, the influenza research and public health communities are strongly
interested in developing a broad-spectrum or universal flu vaccine, which would provide coverage for a
wider range of influenza strains (e.g. all seasonal A/H3N2 strains) or would provide coverage of all
influenza strains (or all influenza A strains), respectively. 1635,1636 Broad-spectrum or universal flu
vaccines would obviate the need for annual production of seasonal flu vaccines and could be used to
protect the public in advance of the next influenza pandemic. Multiple researchers and vaccine production
companies are actively pursuing the development of broad-spectrum or universal flu vaccines. 1637
Demonstrating whether these vaccine candidates are actually drift-resistant, or whether viruses acquire
mutations to escape neutralization by candidate vaccines less readily than to existing vaccines, is a critical
aspect of testing the potential field efficacy of these vaccine candidates.1638
Serial passaging of viruses in cells, in the presence of sera from vaccinated animals, or in vaccinated
animals may lead to the emergence of mutant viruses that can no longer be neutralized by vaccine-
induced antibodies. Sequencing of emergent escape mutants provides insight into how readily viruses can
acquire mutations that confer escape from protective vaccination (i.e. how many mutations are needed to
escape neutralization). Follow-up studies characterizing other properties of emergent escape viruses
relative to the parental virus, such as fitness, may provide additional insight into how likely vaccine
escape mutants are to emerge and persist in human populations. In vitro studies provide a proof of
principle demonstration of whether viruses can mutate to escape vaccines, but virus behavior in response
to relatively simple selection pressures may not translate to human populations. In vivo studies involve
complex selection pressures that more closely mimic those that a virus will encounter during infection of
a vaccinated human host, but results in representative animal models may not translate to human disease.
No alternative approaches are capable of evaluating whether viruses can acquire mutations to escape
neutralization by candidate vaccines prior to field deployment of the vaccine.
vaccines 7: 10-11
vaccines 7: 10-11

Taken together, GoF approaches are uniquely capable of determining whether and how readily influenza
viruses can acquire mutations to escape neutralization by candidate broad-spectrum or universal influenza
vaccines, a critical aspect of testing the potential field efficacy of vaccine candidates.
15.7 Influenza viruses: Detailed analysis of the benefits of GoF research that leads to
evasion of therapeutics
lead to evasion of therapeutics, including licensed therapeutics and therapeutics in development. In this
section, an overview of GoF approaches in this phenotypic category is provided, and the scientific
outcomes and/or products of each approach are described.
15.7.1.1 Serial Passaging of Viruses in the Presence of Therapeutics
Serial passaging of viruses in the presence of a therapeutic may lead to the acquisition of mutations that
allow the virus to evade inhibition by the therapeutic. This approach is performed to determine whether
and how readily a virus evolves resistance in response to selective pressure from a therapeutic and to
identify mutations that confer resistance, which provides a foundation for follow-up studies investigating
the mechanism of action of the therapeutic and the mechanistic basis of antiviral resistance. Follow-up
studies may be performed to determine the consequences of antiviral resistance on other virus
phenotypes, such as fitness. When passaging experiments are performed using a new therapeutic
candidate with an unknown viral target, this information also helps to identify the therapeutic target, as
resistance mutations are most likely to arise in the target protein. Serial passaging approaches have been
performed using cell culture, animal models, and (rarely) human challenge experiments.
15.7.1.2 Forward genetic screen to identify mutations that confer antiviral resistance
Forward genetic screens involve random mutagenesis of antiviral target proteins (e.g. the influenza
neuraminidase protein) followed by screening of mutants to identify those with reduced antiviral
susceptibility (e.g. to NAIs). Follow-up studies may determine the consequences of antiviral resistance
mutations on other virus phenotypes, such as viral fitness. As for serial passaging experiments, the
identification of mutations that confer antiviral resistance provides a foundation for studies to elucidate
antiviral resistance mechanisms.
15.7.1.3 Targeted Modification of Viruses to Introduce Mutations that are Expected to Confer
Antiviral Resistance
A second approach involves targeted genetic modification of a virus to introduce mutations that are
associated with antiviral resistance, which may have been identified through GoF approaches such as
serial passaging or through alt-GoF approaches such as comparative analysis of sequences from patients
who did and did not respond to antiviral treatment. This experiment serves to demonstrate that a particular
mutation or set of mutations is necessary and sufficient to enhance antiviral resistance, which provides a
foundation for follow-up studies investigating the mechanistic basis of antiviral resistance.

Influenza Viruses that are Resistant to Therapeutics
GoF approaches have potential to benefit scientific knowledge by providing insight into the mechanistic
basis of antiviral resistance.
GoF approaches that lead to the identification of mutations that confer antiviral resistance have potential
to inform the interpretation of influenza surveillance data by facilitating the prediction of antiviral
resistance phenotype from genotype, in lieu of isolating and characterizing the antiviral sensitivity of
viruses through phenotypic assays. In the context of seasonal flu surveillance, this application has the
potential to inform therapeutic recommendations for seasonal flu. In the context of animal flu
surveillance, this application has the potential to inform pandemic risk assessments and downstream
decision-making about investments in pandemic preparedness initiatives.
15.7.2.3 Policy decisions
GoF approaches that lead to the identification of molecular markers for antiviral resistance contribute to
assessments of the pandemic risk posed by circulating animal influenza viruses, which are based on
outbreaks. In particular, data about antiviral resistance can inform decisions about whether to pursue
Emergency Use Authorization for new therapeutics in late stages of development, in the event that the
strain under assessment is known or predicted to be resistant to existing antivirals.
GoF approaches that lead to evasion of therapeutics have the potential to benefit the development of
therapeutics in several ways:
• GoF approaches can be used to screen therapeutic candidates based on how readily various
candidates acquire resistance and provide information about whether the emergence of resistance
enhances the transmissibility or virulence of resistant viruses, an important aspect of safety
testing.
• GoF approaches provide information about the potential field efficacy of the therapeutic and the
mechanism of activity of the therapeutic, both of which are critical components of an
• GoF approaches can provide insight into the therapeutic dosing regimens and combination
therapies (e.g. cocktails of monoclonal antibodies) that are the least likely to permit evolution of
resistance.
15.7.2.5 Vaccines
GoF approaches may benefit the production of vaccines through the identification of conserved markers
for neuramidase inhibitor (NAI) resistance. If present in the parental strain upon which a vaccine strain is

based, these markers can be removed from the vaccine virus through targeted deletion or mutagenesis,
which may improve the efficacy and safety of the vaccine production process.
GoF benefits to the development of therapeutics may have downstream economic benefits. Economic
15.7.3 Benefits to Scientific Knowledge
Two classes of antivirals are FDA-approved for general use in the U.S.: the adamantanes, which inhibit
the M2 ion channel,1640 and the neuraminidase inhibitors (NAIs), which inhibit the activity of the NA
protein.1641,1642 As resistance to adamantanes is widespread in seasonal influenza viruses, only NAIs are
recommended for therapeutic use.1643 Three different NAIs are licensed in the US: oseltamivir (Tamiflu®,
FDA-approved in 1999), zanamivir (Relenza®, FDA-approved in 1999), and peramivir (Rapivab®, FDA-
approved for emergency use in 2009 and for general use in 2014). Although most circulating strains have
been sensitive to all NAIs during recent flu seasons, resistance to oseltamivir was widespread during the
2007 – 2008 and 2008 – 2009 seasons, and resistant strains continue to be sporadically detected.1644,1645
Strains that are resistant to oseltamivir or zanamivir as well as strains that are resistant to both drugs have
been observed in nature, in A/H1N1,1646 A/H3N2,1647 and B strains.1648 Resistance has been linked to a
variety of mutations, and in most cases, the mechanisms underlying drug resistance are not well
understood in seasonal flu strains or animal flu strains. In addition, the factors that shape whether resistant
strains will emerge, spread and persist in human populations, including the contribution of viral factors
such as the relative fitness of resistant strains, are unknown.
In this section, the ability of GoF approaches, versus alternative experimental approaches, to address two
unanswered questions in this field are addressed:
1640 Schnell JR, Chou JJ (2008) Structure and mechanism of the M2 proton channel of influenza A virus. Nature 451: 591-595
Chem Soc 119: 681-690
1041

• What are the genetic traits underlying resistance to NAIs, and what is the mechanistic basis of
resistance?
• What selection pressures shape whether and how readily antiviral-resistant strains arise and
spread in nature?
For each question in turn, the potential benefits and limitations of relevant GoF approaches and alt-GoF
approaches are described, then the benefits of GoF approaches relative to alt-GoF approaches are
evaluated. Unique benefits of GoF and alt-GoF approaches are highlighted.
15.7.3.1 Scientific Knowledge Gap 1 – What are the Genetic Traits Underlying Resistance to NAIs, and
Serial passaging of viruses in the presence of one or multiple therapeutics may lead to the emergence of
viruses that are resistant to inhibition by the therapeutic. Sequencing emergent antiviral-resistant viruses
enables the identification of novel mutations that are sufficient to confer resistance. Selection for
resistance studies can be carried out in vitro, in vivo, in animals or through human challenge experiments.
(Human challenge experiments are rare and have only been conducted using human seasonal strains.)
Notably, in vitro and in vivo selection approaches equally enable the identification of mutations
associated with antiviral resistance, though the in vitro approach is faster and cheaper. The in vitro
approach is highly efficient and can be carried out using any virus strain, including currently circulating
strains. Importantly, as multiple mutations may arise during passaging, follow-up studies may be needed
to determine which mutation(s) are responsible for the antiviral resistance phenotype.
Forward genetic screens, which involve random mutagenesis of the NA genes from antiviral-sensitive
strains followed by screening of mutants to identify those with reduced antiviral susceptibility, represent
another GoF approach for discovering novel mutations that confer antiviral resistance. The screening
approach is less efficient that the selection approach but may enable the discovery of rare antiviral
resistance mutations that might be out-competed during a selection experiment due to fitness defects.
Although in principle, this approach could be applied to genes other than NA, to uncover mutations that
confer antiviral resistance through epistatic effects, the relative inefficiency of mutant screens has
practically limited this approach to the NA gene. Depending on the mutagenesis strategy used, follow-up
studies may be needed to determine which mutation(s) are responsible for the antiviral resistance
phenotype. Additionally, for both the serial passaging and forward genetic screen approaches, results may
not translate to other strain contexts.
Targeted genetic modification of parental viruses to introduce mutations associated with antiviral
resistance, followed by phenotypic characterization of the antiviral sensitivity of mutant viruses, is used to
demonstrate that a mutation or set of mutations is necessary and sufficient to confer resistance. Notably,
these mutations may be identified through GoF approaches, such as serial passaging, or alt-GoF
approaches, such as comparative sequence analysis (described below). Subsequently, to determine
whether the phenotypic consequences of mutations are functionally generalizable across multiple virus
strains, targeted mutagenesis can be used to introduce mutations into new virus strains, followed by
characterization of antiviral sensitivity. Together, these results provide a strong foundation for follow-up
biochemical, cell biological, structural, and other studies to determine the mechanistic basis of antiviral
resistance.

Because experiments in this phenotypic category focus on the influenza NA protein, reassortment strains
containing the NA gene or the HA and NA genes from a seasonal strain of interest and the remaining six
or seven genes from the lab-adapted, attenuated strain PR8 (7:1R or 6:2R strains) can be used in lieu of
wildtype seasonal strains for either of the GoF approaches described above. Because these strains are
attenuated relative to the parental strain, this represents one type of alternative approach. Influenza
researchers felt that results about whether mutations do or do not confer antiviral resistance in the context
of attenuated reassortant strains are generally reliable but cautioned that results may not be recapitulated
in the context of the wild type strain.1649 In particular, antiviral resistance mechanisms arising from
reduced NA expression, which has been documented for oseltamivir resistance,1650 or from changes to the
balance of HA and NA proteins expressed on the surface of the virion may function differently in
attenuated reassortant strains. Additionally, 6:2R and 7:1R strains cannot be used to discover or explore
antiviral resistance that arises due to mutations in other virus proteins.
Several alternative experimental approaches can also be used to identify mutations that lead to antiviral
resistance. Comparative sequence analysis of wild type strains that are antiviral-resistant and antiviral-
sensitive enables identification of mutations that are associated with antiviral resistance. However,
because of the high genetic diversity among influenza viruses, identifying relevant mutations may be
difficult. One notable exception is comparative analysis of patient isolates over the course of antiviral
treatment, which is more readily able to identify mutations associated with antiviral resistance due to the
genetic similarity among patient isolates. This approach is most commonly used in immunocompromised
patients due to their longer course of illness.1651,1652,1653,1654 While this approach has successfully identified
mutations associated with oseltamivir and zanamivir resistance, the ability to opportunistically sample
and analyze patient isolates is likely to be relatively rare. In both cases, a causal link between mutations
and antiviral resistance must be established through targeted genetic modification. Re-introduction of
mutations into the parental viruses (GoF) can be used to demonstrate that mutations are necessary and
sufficient to confer antiviral resistance, while deletion of individual mutations from resistant viruses
(LoF) can be used to determine which mutations are necessary for antiviral resistance.
Forward genetic screens to identify mutations that restore antiviral sensitivity to antiviral-resistant strains
(LoF) represents another alternative approach for discovering genetic traits linked to antiviral resistance.
Because this approach involves screening mutants, it is less efficient than GoF approaches for the
discovery of antiviral resistance traits, which rely on selection. Additionally, this approach is limited to
the study of antiviral-resistant strains that have arisen in nature and cannot be used to proactively identify
novel genetic traits that are associated with antiviral resistance. Targeted genetic modification of antiviral-
resistant strains to introduce mutations expected to restore antiviral sensitivity can be used to demonstrate
that a particular trait is necessary for antiviral resistance. Given that single mutations are typically
sufficient to confer resistance to NAIs, targeted LoF and GoF approaches are equally capable of
establishing a causal link between a particular genetic trait and antiviral-resistance. However, use of the
targeted LoF method relies on the existence of an antiviral-resistant strain carrying a particular resistance

(New York, NY) 328: 1272-1275
1651 L'Huillier AG et al (2015) E119D Neuraminidase Mutation Conferring Pan-Resistance to Neuraminidase Inhibitors in an
1653 Gubareva LV et al (1998) Evidence for zanamivir resistance in an immunocompromised child infected with influenza B
virus. J Infect Dis 178: 1257-1262
1654 Kiso M et al (2004) Resistant influenza A viruses in children treated with oseltamivir: descriptive study. Lancet 364: 759-
765

mutation of interest in nature, thus LoF is of limited utility for demonstrating that a resistance trait is
conserved across multiple strain contexts than its GoF counterpart.
The use of in vitro, virus-free systems represents another alternative approach for the study of genetic
traits underlying antiviral resistance. Several in vitro, virus free systems for the study of NAI resistance
have been used, which rely on ectopic expression of the influenza NA gene in cell culture.1655,1656 Using
these systems, forward genetic screens, which involve random mutagenesis of antiviral-sensitive NA
genes of interest followed by ectopic expression of NA mutant libraries and screening for antiviral
resistance, can be used to discover novel mutations that confer resistance. Targeted mutagenesis of wild
type NA genes can then be used to demonstrate that a particular mutation or set of mutations is necessary
and sufficient to confer resistance, as well as to determine whether the phenotypic consequences of the
mutation(s) are conserved across multiple genetic contexts. This approach can be successfully used to
study mutations that confer resistance by altering the function of the NA protein. However, this approach
cannot be used to uncover or to study mutations that confer resistance by altering the expression levels of
the NA protein, as has been documented for the H274Y mutation (N1 numbering),1657 or mutations in
other genes that give rise to resistance through epistatic effects. Additionally, given that antiviral-
resistance is a continuum, results may not be recapitulated (or be clinically relevant) in the context of the
full virus.
Finally, computational models have been used to predict mutations that disrupt binding between NAIs
and the NA protein, which are expected to lead to antiviral resistance. While these models can be used to
generate hypotheses about antiviral resistance mutations in any virus strain, all predictions must be
experimentally confirmed through targeted mutagenesis, a GoF approach. Additionally, this method
cannot be used to predict mutations that give rise to resistance through other mechanisms.
15.7.3.1.3 Summary – Benefits of GoF Approaches Relative to Alt-GoF Approaches
into the mechanisms underlying viral resistance to NAIs. Taken together, GoF approaches are uniquely
capable of identifying mutations that are necessary and sufficient to confer antiviral resistance across
multiple strain contexts, which provides a strong foundation for follow-up studies to elucidate the
mechanisms underlying antiviral resistance. GoF approaches also represent the most efficient and
effective approach for discovering novel mutations that confer antiviral resistance in any virus strain, as
conducting experiments with wild type viruses allows for discovery of the full spectrum of mutations that
may confer resistance, including mutations that alter the function or expression level of the NA gene as
well as mutations in other virus proteins that cause resistance through epistatic effects. Attenuated
reassortant strains may be used in lieu of wild type strains for many of these experiments, but results may
not be recapitulated in the context of the wild type viruses, particularly if antiviral resistance arises
through mechanisms other than changes to the function of the NA protein.
Alternative approaches can provide valuable insight into the study of antiviral resistance mechanisms but
have limitations relative to GoF approaches. Discovering new genetic traits associated with antiviral
resistance through comparative analysis of wild type sequences may be difficult. The comparison of
patient isolates over the course of antiviral treatment is a notable exception, but opportunities for such
studies are likely to be relatively rare. LoF approaches are relatively inefficient for the discovery of novel
1655 Nivitchanyong T et al (2011) Enhanced expression of secretable influenza virus neuraminidase in suspension mammalian
cells by influenza virus nonstructural protein 1. Journal of virological methods 178: 44-51
1656 Schmidt PM et al (2011) A Generic System for the Expression and Purification of Soluble and Stable Influenza
Neuraminidase. PLoS ONE 6: e16284
(New York, NY) 328: 1272-1275

genetic traits associated with antiviral resistance but can be used to demonstrate that a particular mutation
is necessary for antiviral resistance. Notably, the targeted LoF approach is often as capable of establishing
a causal link between a particular mutation and antiviral resistance as the targeted GoF approach because
NAI resistance is often conferred by single mutations; however, the ability of targeted LoF to demonstrate
that particular markers are conserved across strain contexts is limited by the number of of antiviral
resistant strains in nature. In vitro virus-free systems can be used to discover and validate mutations in the
NA gene that give rise to resistance but are not suitable for the study of resistance mechanisms that
involve alterations to gene expression levels or epistatic effects, and results may not be recapitulated in
the context of the full virus. Computational models may be used to predict novel mutations that confer
resistance by disrupting binding between the NAI molecule and the NA protein, but all predictions must
be experimentally confirmed using GoF approaches.

Table 15.32: Summary of the Benefits of GoF Approaches that Lead to Evasion of Therapeutics
Scientific Knowledge Benefits – What are the Genetic Traits Underlying Resistance to NAIs, and What is the Mechanistic Basis of Resistance?
• Identify novel mutations that are sufficient

GoF #1 [1]*: to confer antiviral resistance phenotype
In vitro approach: serial passaging of • Approach is highly efficient
antiviral-sensitive virus in cells in the • Proactive – can be carried out using any
presence of antiviral virus strain, including strains that have not • Associative – Information produced is correlative, not
yet gained resistance in nature causative
• Identify novel mutations that are sufficient • Narrow breadth – Results may not generalize to other
GoF #2 [2]: virus strains
to confer antiviral resistance phenotype
In vivo approach: Serial passaging of
antiviral-sensitive virus in animals in the • Proactive – can be carried out using any
presence of antiviral virus strain, including strains that have not
yet gained resistance in nature
GoF #3 [3]:
“Passaging in humans” – human challenge
• Identify mutations that are associated with • Associative – Information produced is correlative, not
experiments
antiviral resistance in vivo causative
• Challenge human volunteers with drug- • Ethical considerations limit the number of experiments
sensitive influenza strains and treat with
• Proactive – can be carried out using any
virus strain, including strains that have not that can be carried out
antivirals
yet gained resistance in nature • Narrow breadth – Results may not generalize to other
• Compare virus sequences from isolates virus strains
collected over the course of antiviral
treatment

• Identify novel mutations that are necessary

and sufficient to confer antiviral resistance
phenotype • Narrow breadth - Results may not generalize to other
virus strains
GoF #4 [4]: • Ability to identify rare mutations that may
Forward genetic screen to identify mutations be out-competed during selection • Forward genetic screen to identify novel antiviral
that confer antiviral resistance experiments resistance markers is inefficient relative to GoF
approaches; practically limited to investigating mutations
• Proactive – can be carried out using any in the NA protein
virus strain, including strains that have not
yet gained resistance in nature
• Identify mutations that are necessary and

sufficient to confer antiviral resistance
phenotype
GoF #5 [5, 6]:

• Gain insight into mechanisms underlying
antiviral resistance • Lack of publication of negative results – Compromises
Targeted genetic modification of antiviral-
• Proactive – can be carried out using any evaluation of whether the function of particular markers is
sensitive virus to introduce mutation(s)
virus strain, including strains that have not broadly conserved
associated with anti-viral resistance.
yet gain resistance in nature
• Enables testing of markers in different
strain contexts to assess generalizability of
previous findings

causative
Alt-GoF #1 [1]:
“Passaging in humans” – comparative • Narrow breadth – Results may not generalize to other
• Identify mutations that are associated with virus strains
sequence analysis of patient isolates from
evolution of antiviral resistance in vivo
multiple time points over the course of • Limited patient availability constrains the number of
antiviral treatment studies that can be done
• Limited to studying strains that infect study subjects


causative
Alt-GoF #2 [2]:
• Identify mutations that are associated with • Limited by the quality and availability of surveillance
Characterization of wildtype viruses antiviral resistance data
Comparative sequence analysis of natural
antiviral-resistant and antiviral-sensitive
• High genetic diversity impairs identification of relevant
mutations
virus strains
• Reactive - Limited to study antiviral resistant strains that
have already arisen in nature
• Forward genetic screen to identify novel antiviral

resistance markers is inefficient relative to GoF
Alt-GoF #3 (loss of function) [3, 8, 9]: approaches; practically limited to investigating mutations
in the NA protein
• Forward genetic screen to identify • Identify novel mutations that are necessary
mutations that decrease antiviral for antiviral resistance
• Narrow breadth - Results may not generalize to other
resistance virus strains
• Gain insight into mechanisms underlying • Reactive – Limited to studying antiviral resistant strains
• Targeted mutagenesis of antiviral- antiviral resistance
resistant strains to introduce mutations that have already arisen in nature
expected to confer antiviral sensitivity • Limited utility for demonstrating functional
generalizability of particular markers across multiple
strain contexts
Alt-GoF #5 [5, 6, 10]:

Use of in vitro, virus free systems:
• Identify mutations that are necessary and • Limited to studying resistance mutations/mechanisms that
• Forward genetic screen to identify sufficient to confer antiviral resistance
mutations that increase antiviral resistance involve altering the function of the NA protein
• Enables testing of markers in different • Simplicity of model system: Results may not be
• Targeted genetic modification of strain contexts to assess generalizability of
antiviral-sensitive NA gene to introduce recapitulated in the context of the full virus
previous findings
mutations expected to confer antiviral
resistance

• Predictive – Does not confirm or correlate phenotypic

Alt-GoF #6 [7, 11]: effects in a biological context
In silico: computer modeling to predict • Predict mutations that are necessary and
mutations that disrupt binding between NAIs sufficient to disrupt antiviral binding to its • Model accuracy – Utility of the approach depends on the
and the NA protein target protein, which are expected to confer quality of existing models
resistance
• Limited to studying resistance mechanisms that involve
disruption of NAI-NA interaction
information).

15.7.3.2 Scientific Knowledge Gap 2 – What Selection Pressures Shape Whether and How Readily
Serial passaging of viruses in the presence of one or more therapeutics to select for antiviral-resistant
strains provides insight into whether, how readily, and how antiviral resistance arises in response to
selective pressure from therapeutics. These experiments have been conducted in vitro and in vivo, through
animal experiments and human challenge experiments. Due to the simple selection pressures encountered
by viruses during passage in cell culture, the in vitro approach is less useful than the in vivo approach for
understanding how selection pressures in humans are likely to drive the emergence of antiviral-resistant
viruses. The ability to gain direct insight into emergence of resistance in humans through human
challenge experiments is valuable, but ethical considerations severely constrain the number and scope of
experiments that can be carried out. Additionally, variability in host factors (e.g. past exposure to
influenza viruses) may complicate interpretation of findings. Animal experiments provide a controlled
system for studying the emergence of resistance under complex selection pressures, including identifying
resistance mutations that arise but are negatively selected within or between hosts. However, results may
not translate to human populations.
Additionally, characterizing the fitness, infectivity, and transmissibility of antiviral-resistant viruses

generated through GoF approaches, including serial passaging and targeted mutagenesis, may provide
insight into how likely resistant viruses are to emerge, spread, and persist in human populations. In
particular, the targeted mutagenesis approach provides a controlled system for studying the interplay
between antiviral resistance and other virus properties by enabling comparison of genetically similar
viruses that differ only (or primarily) in their antiviral sensitivity.
Several alternative approaches can be used to gain insight into selection pressures that shape the evolution
and spread of antiviral resistance. Comparative analysis of the sequences and phenotypic characteristics
of patient isolates over the course of antiviral treatment has potential to provide direct insight into the
mechanisms driving emergence of antiviral resistance in people, including identifying resistance
mutations that are negatively selected. However, as these studies are typically conducted in
immunocompromised patients due to their longer course of illness, results may not be representative of
the general population. In addition, relative to animal passaging experiments (GoF), opportunities to
conduct studies involving patients are likely to be relatively rare due to ethical considerations.
Comparative analysis of the phenotypic properties (e.g. fitness) of antiviral-resistant and antiviral-
sensitive wild type strains can reveal genetic and phenotypic changes that are associated with the
acquisition of antiviral resistance (including associations between antiviral resistance and other virus
properties), which may provide insight into the viral properties that shape the evolution and spread of
antiviral resistance in nature. However, this approach has significant limitations. First, the surveillance
record is static and cannot provide insight into negatively selected traits. Second, current surveillance
efforts, which largely involve consensus sequencing, are unlikely to capture the emergence of rare
antiviral-resistant variants. Finally, due to the high genetic diversity among influenza viruses, this
approach cannot establish a causal link between the acquisition of antiviral resistance and other
phenotypic changes (e.g. changes in viral fitness). For these reasons, comparative analysis of wild type
viruses provides limited insight into the evolutionary mechanisms shaping the evolution and spread of
antiviral resistance in nature.

Similar to GoF targeted mutagenesis approaches, targeted LoF approaches (i.e. targeted genetic
modification of antiviral-resistant strains to introduce mutations that restore antiviral susceptibility)
provides a controlled system for studying the interplay between antiviral resistance and other virus
phenotypes, such as fitness.
Other alternative approaches are not suitable for the study of evolutionary pressures that shape the
emergence and spread of antiviral resistance. In vitro, virus free approaches cannot provide insight into
how antiviral resistance affects other virus phenotypes, and current computational models cannot account
for epistatic effects (e.g. how antiviral resistance affects fitness). The use of attenuated reassortant strains
for GoF selection approaches, in lieu of wild type viruses, is of limited utility for studying the evolution
of antiviral resistance because the fitness of attenuated strains is altered relative to the wild type strains.
15.7.3.2.3 Summary – Benefits of GoF Approaches Relative to Alt-GoF Approaches
into the evolutionary mechanisms underlying acquisition and spread of viral resistance to NAIs. Taken
together, GoF approaches, namely serial passaging of viruses in animals in the presence of therapeutics,
represent the most efficient and effective strategy for gaining in-depth insight into the viral and host
selection pressures that shape the emergence and spread of antiviral resistance. Notably, attenuated
reassortant strains cannot be used for these studies because the phenotypic properties that are likely to
shape the likelihood that antiviral resistant strains will spread and persist in human populations, such as
fitness, are altered in these strains. While gaining direct insight into the behavior of the virus in humans
through human challenge studies (GoF) is valuable, these studies are rare due to ethical considerations
and interpretation of findings is complicated by variability in host factors, such as past exposure to
influenza. Comparative analysis of patient isolates over the course of antiviral treatment can also provide
in-depth insight into the evolution of antiviral resistance in people, but studies are typically conducted in
immunocompromised patients and thus may not translate to healthy populations. Comparative analysis of
wild type isolates provides limited mechanistic insight into the viral or host factors that shape evolution of
antiviral resistance. Finally, neither virus-free approaches nor in silico approaches can be used to study
the interplay between antiviral resistance and other virus phenotypes.

Scientific Knowledge Benefits – What Selection Pressures Shape Whether and How Readily Antiviral-Resistant Strains Arise and Spread in Nature?
• Provide in-depth insight into whether and how • Translatability – Results in cell culture systems
readily antiviral resistance arises, and the
may not translate to human populations
underlying evolutionary mechanisms
GoF #1 [1]*: • Proactive – can be carried out using any virus • Simplicity of selection pressures in vitro render
this approach less useful than the in vivo
In vitro approach: serial passaging of antiviral- strain, including strains that have not yet gain
approach
sensitive virus in cells in the presence of antiviral resistance in nature
• Gain associative insight into the interplay • Associative – Does not establish a causal link
between antiviral resistance and other virus
phenotypes
phenotypes, such as fitness
• Provide in-depth insight into whether and how

readily antiviral resistance arises, and the
GoF #2 [2]: underlying evolutionary mechanisms • Translatability – Results in animal models may
not translate to human populations
In vivo approach: Serial passaging of antiviral- • Proactive – can be carried out using any virus
sensitive virus in animals in the presence of strain, including strains that have not yet gain • Associative – Does not establish a causal link
antiviral resistance in nature between antiviral resistance and other virus
phenotypes
• Gain associative insight into the interplay
GoF #3 [3]: • Provide in-depth insight into whether and how

readily antiviral resistance arises in humans, • Ethical considerations limit the number of
“Passaging in humans” – human challenge
and the underlying evolutionary mechanisms experiments that can be carried out
experiments
• Proactive – can be carried out using any virus • Variability in host factors may complicate
• Challenge human volunteers with drug- strain, including strains that have not yet gain interpretation of findings
sensitive influenza strains and treat with
resistance in nature
antivirals • Associative – Does not establish a causal link
• Gain associative insight into the interplay between antiviral resistance and other virus
• Compare virus sequences from isolates between antiviral resistance and other virus phenotypes
collected over the course of antiviral treatment

• Gain insight into the interplay between

antiviral resistance and other virus phenotypes,
such as fitness
GoF #4 [5, 6]:
• Controlled system – enables comparison of • Translatability – Results in cell culture or animal
Targeted genetic modification of antiviral-
genetically similar viruses that differ only in
sensitive virus to introduce mutation(s) associated models may not translate to human populations
their antiviral sensitivity phenotype
with anti-viral resistance.
• Proactive – can be carried out using any virus
strain, including strains that have not yet gain
resistance in nature
• Limited patient availability constrains the number

of studies that can be done
• Provide in-depth insight into whether and how • Results from studies involving
Alt-GoF #1 [1]: readily antiviral resistance arises in humans, immunocompromised patients (common) may not
“Passaging in humans” – comparative sequence and the underlying evolutionary mechanisms be representative of the general population
analysis of patient isolates from multiple time • Gain insight into the interplay between • Limited to studying strains that infect study
points over the course of antiviral treatment antiviral resistance and other virus phenotypes, subjects
such as fitness
• Associative – Does not establish a causal link
phenotypes

• Static – Cannot identify lost or negatively

selected traits
• Limited by the quality and availability of
Alt-GoF #2 [2]: • Gain associative insight into the interplay surveillance data
Characterization of wildtype viruses between antiviral resistance and other virus • Consensus sequencing is unlikely to capture the
Comparative sequence analysis of natural phenotypes, such as fitness emergence of rare antiviral-resistant strains
antiviral-resistant and antiviral-sensitive virus • Associative – Does not establish a causal link
strains between antiviral resistance and other virus
phenotypes
• Reactive – Limited to studying antiviral resistant
strains that have already arisen in nature
• Gain insight into the interplay between

antiviral resistance and other virus phenotypes,
such as fitness
• Controlled system – enables comparison of • Translatability – Results in cell culture or animal
Alt-GoF #3 [3, 8, 9]: genetically similar viruses that differ only in models may not translate to human populations
Targeted mutagenesis of antiviral-resistant strains their antiviral sensitivity phenotype
to introduce mutations expected to confer
• Reactive – Limited to studying antiviral resistant
strains that have already arisen in nature
antiviral sensitivity (loss of function)
• Identify novel mutations that are necessary for
antiviral resistance
• Gain insight into mechanisms underlying
antiviral resistance
information).

15.7.4 GoF Benefits to Surveillance
Two classes of antivirals are FDA-approved for general use in the U.S.: the adamantanes, which inhibit
the M2 ion channel,1658 and the neuraminidase inhibitors (NAIs), which inhibit the activity of the NA
protein.1659,1660 As resistance to adamantanes is widespread in seasonal influenza viruses, this class of drug
is no longer recommended for therapeutic use. Through influenza surveillance, public health professionals
monitor the appearance and prevalence of NAI-resistant strains of seasonal influenza viruses circulating
in global populations and of animal influenza viruses that have caused human infections.1661 Data on the
prevalence of antiviral-resistant seasonal strains informs therapeutic recommendations developed by the
CDC (i.e. which of the three FDA-approved NAIs should be recommended as a first-line treatment).1662 In
the context of surveillance for zoonotic influenza infections in humans, this data informs decision-making
about pandemic preparedness initiatives because antiviral resistance is one of the risk elements considered
in a pandemic risk assessment.
fluorometric 20-(4-methylumbelliferyl)-a-D-N-acetylneuraminic acid (MUNANA) assay or other assays
and/or by inspecting sequences for the presence of mutations that are known to confer NAI resistance.
GoF approaches can improve the practice of using molecular markers by enabling the discovery of new
antiviral resistance markers and by validating known markers in new strain contexts. This section first
reviews these GoF benefits relative to alternative experimental approaches that may provide similar
information. Subsequently, the utility of laboratory assays versus sequence-based prediction for
characterizing the antiviral sensitivity of surveillance isolates is analyzed. The public health actions that
are taken downstream of this assessment are described in section 16.7.5, below.
15.7.4.1 Using Molecular Markers for Antiviral Resistance to Interpret Surveillance Data
15.7.4.1.1 Current utility and shortcomings of using molecular marker data to predict the antiviral
sensitivity phenotype of viruses detected through surveillance
Many mutations have been identified that confer resistance to one or multiple NAIs. In part because NAI
resistance can arise from one or two mutations, the predictive value of these markers is generally much
stronger than that of markers associated with adaptation, transmissibility, and virulence, which are the
result of a constellation of genetic changes.1663,1664,1665 Multiple markers for NAI resistance have been
shown to be functionally generalizable, conferring resistance in multiple strain contexts.1666 In the
experience of influenza researchers and government officials involved in surveillance, the presence of a
1658 Schnell JR, Chou JJ (2008) Structure and mechanism of the M2 proton channel of influenza A virus. Nature 451: 591-595
Chem Soc 119: 681-690
1662 Ibid.
1664 Sleeman K et al (2013) R292K substitution and drug susceptibility of influenza A(H7N9) viruses. Emerging infectious
diseases 19: 1521-1524
1665 Boivin G (2013) Detection and management of antiviral resistance for influenza viruses. Influenza and Other Respiratory
Viruses 7: 18-23
1666 Ibid.

validated antiviral resistance marker is strongly predictive for antiviral resistance, though all agreed that
sequence-based predictions must be experimentally confirmed. However, the absence of a known
resistance marker is not necessarily predictive of antiviral sensitivity, as it is likely that additional
mutations or sets of mutations can lead to resistance, in particular to multi-drug resistance. This lack of
knowledge about the mutational landscape that permits evolution of antiviral resistance limits the utility
of sequence-based approaches for predicting resistance. Moreover, validating known markers in
additional strain contexts will further strengthen their predictive value. As discussed in detail in section
16.7.3.1 above, both GoF and alt-GoF approaches can provide insight into these scientific questions. The
relevant findings are summarized here.
15.7.4.1.2 Summary - Benefits of GoF Approaches Relative to Alt-GoF Approaches for Improving the
Utility of Molecular Markers for Antiviral Resistance
GoF approaches represent the most efficient and effective strategy for discovering novel mutations that
give rise to antiviral resistance and are uniquely capable of confirming that particular mutations are
necessary and sufficient to confer resistance in multiple strain contexts. Notably, for mutations that confer
resistance by altering the function of the NA protein (i.e. versus altering NA expression levels or through
epistatic effects), these experiments can be performed using attenuated reassortant strains, though results
may not be recapitulated in the context of the wild type strain. In vitro, virus-free systems can also be
used to discover and validate new mutations that give rise to antiviral resistance by altering the function
of the NA protein, but results should be confirmed in the context of the full virus. Given that single
mutations are sufficient to confer resistance to NAIs, targeted mutagenesis of antiviral-resistant strains to
restore antiviral sensitivity (LoF) is also capable of establishing a causal link between a particular trait
and antiviral resistance. However, because this approach relies on the existence of antiviral resistant
strains in nature, the ability of this approach to demonstrate that particular markers are conserved across
multiple strain contexts is limited relative to GoF approaches. Comparing the sequences of wild type
viruses or of patient isolates over the course of antiviral treatment may lead to the identification of
mutations that are associated with antiviral resistance, but all hypotheses must be confirmed using
targeted mutagenesis (GoF or LoF) to be useful for surveillance. In addition, these approaches are limited
to the discovery of antiviral resistance mutations that have already arisen in nature. Computational models
can be used to predict mutations that disrupt the interaction between an NAI compound and an antiviral,
but predictions must be validated experimentally. Taken together, GoF approaches provide unique
benefits for strengthening the predictive value of molecular markers for antiviral resistance, thereby
improving their utility for interpreting surveillance data.
Of note, NAI resistance markers that have been shown to be conserved across multiple strain contexts and
are currently incorporated into the practice of analyzing surveillance data.1667 Thus, the benefits of GoF
research about molecular markers for antiviral resistance to the practice of surveillance can be realized
immediately.
15.7.4.1.3 Utility of Molecular Markers for Antiviral Resistance in Surveillance, Relative to Phenotypic
Assays
MUNANA assay or other phenotypic assays and/or by inspecting sequences for the presence of mutations
that are known to confer NAI resistance. For characterizing surveillance isolates, both methods have
strengths and limitations.
1667 (2015u) Interviews with influenza researchers and U.S. government representatives involved in influenza surveillance.

The strength of phenotypic assays, relative to predictive approaches, is that phenotypic assays provide a
direct readout of antiviral resistance. However, the practice of characterizing the antiviral sensitivity of
surveillance isolates through phenotypic assays has several shortcomings. These shortcomings were
discussed in detail in section 15.3.4 and are briefly summarized here. First, the need for viral isolates
limits the number of viruses that can be subjected to phenotypic characterization. Second, the composition
of viral species present in the original clinical sample changes during isolation, as the most fit viral quasi-
species outcompete others. This change is of particular concern for antiviral resistance testing because
antiviral-resistant viruses are often less fit than antiviral-sensitive viruses, thus the presence of antiviral
resistant strains in mixed infections can be obscured by viral isolation. One government official involved
in the pandemic risk assessment process reported that such mixed infections do occur; in one case, the
results of antiviral resistance assays were indeterminate, while sequencing of the clinical isolate revealed
the presence of both viral genotypes.1668 Finally, in the event that clinical samples or viral isolates are
shipped to WHOCCs for antiviral susceptibility testing, delays stemming from logistical, political, and/or
regulatory factors create a lag time between sample collection and phenotypic characterization.
The practice of predicting the antiviral resistance phenotype of surveillance viruses based sequence
inspection for molecular markers of antiviral resistance addresses several shortcomings associated with
the phenotypic assay approach. In particular, the fact that clinical isolates can be directly sequenced
provides several advantages. First, this method provides a direct readout of the viral species present in the
sample, avoiding the problem that the composition of viral quasispecies changes during the virus isolation
process. Second, following inactivation of virus present in a clinical sample, the sequencing procedure
can be carried out under BSL-2 conditions and thus can more feasibly be implemented at NICs and other
diagnostic labs in developing countries. Third, whether from clinical samples or virus isolates, sequencing
is becoming ever cheaper and easier. As a result, viral genetic sequence data is currently the fastest and
most reliable data generated by diagnostic labs in areas where viruses of concern are circulating.1669
However, most genetic surveillance data is generated by sequencing of viral isolates at WHOCCs, though
the number of NICs with sequencing capabilities as well as the number of diagnostic labs (including
NICs, WHOCCs, and collaborating labs) that conduct sequencing on clinical samples is increasing. Full
realization of the benefits that can be derived from the use of molecular marker data will require an
expansion of sequencing capabilities at diagnostic laboratories that comprise the “base” of the influenza
surveillance system as well as an increase in the number of clinical samples that are directly sequenced.
The major limitation of using molecular markers is that the predictive value of molecular markers for
antiviral resistance is sub-optimal, in part due to a lack of knowledge of the mutational landscape that can
give rise to antiviral resistance and in part due to limited knowledge about whether certain markers will
convey antiviral resistance in new strain contexts. GoF approaches have potential to address both of those
questions, which will improve the utility of molecular marker data to surveillance. Given that several
markers have already been shown to be conserved across multiple strain contexts and that only one or two
mutations are needed to confer NAI resistance, additional conserved markers for NAI resistance are likely
to be identified. Whether similarly conserved antiviral resistance markers can be identified for future
antivirals is unknown. Similar to the NAIs, single mutations have been shown to confer resistance to
multiple antivirals in development, but the genetic threshold for resistance to some future antivirals may
be higher, in which case sequence-based predictions of antiviral resistance become more difficult. Finally,
given the inherent uncertainty of sequence-based predictions, researchers and governmental officials
involved in the analysis of surveillance data emphasized that predictions should be validated through
antiviral resistance assays whenever possible.
1668 (2015t) Interview with BARDA representative.


Taken together, for characterizing the antiviral sensitivity or surveillance isolates, both phenotypic assays
and inspection of sequences for molecular markers of antiviral resistance have strengths and limitations
(summarized in Table 15.34). Phenotypic assays provide direct information about the degree of antiviral
resistance of a particular strain, but results are delayed relative to sample collection and the properties of
viral isolates may not reflect the properties of viral quasispecies present in the original clinical sample.
For these reasons, researchers and government officials involved in influenza surveillance value the
ability to corroborate phenotypic assay data with sequence-based predictions based on molecular markers
of antiviral resistance, particularly when clinical samples can be directly sequenced. As sequencing
becomes more common at NICs and other diagnostic laboratories and the number of known, validated
markers for NAI resistance rises, the molecular marker approach will take on relatively greater
importance. Ultimately, due to the rapidity of sequence-based analysis relative to phenotypic assays, the
use of molecular markers may increase capacity to monitor for antiviral resistance.

Surveillance benefits – Aid evaluation of the antiviral susceptibility of circulating seasonal and animal influenza viruses
• Improves the accuracy of antiviral susceptibility information • The predictive value of markers for antiviral
o Clinical samples can be directly sequenced resistance is currently sub-optimal
o Corroboration of phenotypic assay results increases the robustness of o Scientific knowledge about the landscape of
the data mutations that can give rise to antiviral
GoF #1: • Increases the quantity and timeliness of antiviral susceptibility resistance is incomplete
Inspection of sequences for information • The use of molecular markers is inherently
the presence of molecular o As sequencing becomes cheaper and easier, a greater number of predictive
markers for antiviral viruses may be sequenced than subjected to phenotypic assays o Predictions should be validated through
resistance o Enables rapid evaluation of antiviral susceptibility, in the event that phenotypic testing whenever possible
sequencing data is generated at the site (or in the country) of sample • Full realization of benefits depends on
collection expanding sequencing capabilities at NICs and
• Molecular marker data are currently used to interpret surveillance data increasing the number of clinical samples that
• New data can be incorporated into the process in the immediate term are directly sequenced
• The need for viral isolates limits the number

of viruses that can be characterized
• The composition of viruses present in the
Alt-GoF #1: original clinical samples may change during
Phenotypic • Provides a direct readout of antiviral susceptibility virus isolation
characterization of the o In mixed infections with antiviral-sensitive
antiviral susceptibility of and –resistant species, less fit resistant
surveillance isolates species may be out-competed during the
isolation process
• Sample shipping delays due to logistical,
political, and regulatory factors delay the
generation of phenotypic data

15.7.5 Benefits to Decision-Making in Public Health Policy
GoF approaches have potential to benefit surveillance for antiviral resistant strains by improving the
practice of using molecular markers for antiviral resistance to infer antiviral resistance from genotype.
Surveillance for antiviral resistant strains informs downstream decision-making related to public health
practice and policy. Namely, data on the prevalence of antiviral-resistant seasonal strains informs
therapeutic recommendations developed by the CDC, and antiviral resistance is one of the risk elements
considered in a pandemic risk assessment of an animal influenza virus. This section describes each of
these applications, which illustrates the ultimate public health impacts associated with GoF benefits to
surveillance. Alternative data sources that inform these public health decisions are also evaluated.
15.7.5.1 Benefits to Decision-Making Related to Seasonal Flu Strains
Two classes of antivirals are FDA-approved for general use against seasonal influenza strains: the
adamantanes, which are no longer recommended for therapeutic use due to widespread resistance, and the
NAIs, which includes three different small molecule drugs (oseltamivir, zanamivir, and peramivir). 1670,1671
recommendations to clinicians for the use of influenza antivirals. For example, the adamantane class of
influenza antivirals (M2 inhibitors) were recommended until 2005, when widespread resistance (>90%)
was detected among strains circulating during the 2005 -2006 flu season. This triggered CDC to issue an
interim change in their antiviral treatment guidelines, recommending the use of NAIs in lieu of
adamantanes.1672 Although NAI resistance was high during the 2007 – 20081673 and 2008 – 20091674
seasons (>98% of H1N1 isolates tested), recent outbreak strains have remained susceptible to all three
NAIs. However, seasonal strains that are resistant to one and to multiple NAIs have been detected in
nature,1675 sporadic cases of oseltamivir-resistant 2009 H1N1 pandemic virus continue to be detected, and
development of resistance to oseltamivir or zanamivir during treatment of seasonal influenza has been
documented.1676,1677,1678 Current antiviral treatment guidelines do not recommend particular NAIs;
however, an increase in the prevalence of singly-resistant strains could trigger a recommendation change.
As antivirals are most effective when given within 48 hours of symptom onset, the CDC recommends
initiating antiviral treatment prior to laboratory confirmation of influenza (i.e. without knowledge of
antiviral susceptibility).1679 Given that, antiviral treatment recommendations based on reliable knowledge
about the prevalence of resistance to particular antivirals among circulating strains is essential for the
success of therapeutic treatment. Currently, a subset of the influenza viruses that are collected by
1671 (2006) High levels of adamantane resistance among influenza A (H3N2) viruses and interim guidelines for use of antiviral
agents--United States, 2005-06 influenza season. MMWR Morb Mortal Wkly Rep 55: 44-46
1672 ibid.
1041
1675 L'Huillier AG et al (2015) E119D Neuraminidase Mutation Conferring Pan-Resistance to Neuraminidase Inhibitors in an
oseltamivir. Ibid. 183: 523-531
1677 Kiso M et al (2004) Resistant influenza A viruses in children treated with oseltamivir: descriptive study. Lancet 364: 759-
765
1678 Hurt AC et al (2009) Zanamivir-Resistant Influenza Viruses with a Novel Neuraminidase Mutation. J Virol 83: 10366-
10373

WHOCCs are sent to CDC for antiviral susceptibility testing.1680 As discussed above, phenotypic assay
results are often corroborated by sequence inspection for the presence of molecular markers associated
with antiviral resistance. As sequencing becomes more common at NICs and other diagnostic laboratories
and the number of known, validated markers for NAI resistance rises, the rapidity of the molecular
marker approach may expand the number of surveillance viruses that can be phenotypically characterized.
This expansion will provide a stronger foundation for antiviral treatment recommendations and may
enhance detection of rare antiviral-resistant variants to increase awareness.
15.7.5.2 Benefits to Decision-Making Related to Pandemic Influenza
preparedness initiatives (discussed in detail in section 16.3.5.2). This section analyzes the value of
antiviral resistance information relative to other types of information considered in risk assessments, as
well as particular contributions of antiviral resistance information to downstream decision-making.
For pandemic risk assessments, the antiviral resistance risk element does not contribute to the likelihood
that an animal virus will emerge to efficiently infect and transmit in humans and moderately contributes
to the assessment of the expected consequences of an emergence event. For example, in a recent risk
assessment of avian H7N9, avian H1N1, and swine H3N2v viruses, the antiviral resistance element was
worth less than the disease severity, population immunity, and extent of human infections risk elements
(approximately one-third as much as the most highly weighted disease severity element).1681 Stakeholders
involved in the pandemic risk assessment process emphasized that antiviral resistance does not increase
risk a priori but rather is important when coupled to other factors indicative of increased pandemic
potential, such as a high number of human infections or enhanced transmissibility or virulence in ferrets.
In this case, the observation of antiviral resistance may trigger USG representatives to initiate the process
of applying for Emergency Use Authorization (EUA) from the FDA for antivirals in development, to
ensure that effective antivirals will be available if the strain under evaluation spreads to cause a pandemic.
Importantly, when evaluating antiviral resistance, stakeholders consider both phenotypic and genetic data,
given the caveats associated with both types of data (discussed above). Additionally, the ability to
conduct a rapid risk assessment based on sequence data, by inspection of sequences for molecular
markers of antiviral resistance, is valuable when strains first emerge and sequences are published prior to
the receipt of viral isolates. For example, in 2013, the observation that the sequences of early clinical
isolates of avian influenza H7N9 in China contained molecular markers previously shown to confer to
oseltamivir and zanamivir informed BARDA’s decision to move forward with the decision to initiate the
EUA process for antivirals in development.1682 Given the two-week lag time between publication of the
H7N9 sequence and additional time needed to conduct antiviral sensitivity testing, the ability to use
molecular markers to infer antiviral resistance phenotypic from genotype provided a several week head
start on the EUA process. There is no set time frame for approval of an EUA, but approval can be granted
within days if the FDA has already reviewed the relevant data on the MCM (submitted in advance as a
“pre-EUA” package).1683 For example, the FDA issued an EUA for peramivir in October 2009 in response
to the H1N1 influenza pandemic three days after the request, and recently issued an EUA for the DoD
EZ1 rRT-PCR Ebola diagnostic in August 2014 one day following the request. In both cases, the FDA
had worked with their government partners on pre-EUA packages in advance of the requests. Thus, a
1682 (2015t) Interview with BARDA representative.

several week head start on the process could significantly impact the timing of availability of antivirals in
the event of a pandemic.1684
Taken together, this suggests that the ability of GoF benefits to surveillance for antiviral-resistant viruses
to contribute to the overall risk assessment score is moderate and that the ability the infer antiviral
resistance phenotype based on genotype may provide a valuable head start on the EUA process for
antivirals in development when novel strains that are resistant to licensed antivirals emerge.
15.7.6 GoF Benefits to the Development of vaccines
15.7.6.1 Vaccine Development Benefit: Targeted Mutagenesis to Remove Antiviral Resistance Markers
from Vaccine Viruses
Vaccine viruses comprise the HA and NA genes from the wild type strain of interest and the remaining
six genes from a vaccine backbone virus such as PR8. Mutations that confer resistance to NAIs, the one
approved class of influenza antivirals that are recommended for use in the U.S., arise in the NA
gene.1685,1686,1687,1688 If the wild type NA gene contains conserved markers for NAI resistance, these
markers can be removed through targeted deletion or mutagenesis to increase the safety of the vaccine
production process. (Of note, most influenza vaccines produced in the U.S. are inactivated, thus whether a
vaccine strain is sensitive or resistant to antivirals has no impact on the safety of the vaccine itself.) For
example, this strategy was used for production of a pre-pandemic H7N9 vaccine in 2013.1689 By sequence
inspection, clinical isolates from the first few cases of H7N9 were found to contain the R292K mutation,
which had previously been shown to reduce resistance to multiple NAIs.1690 As a result, this mutation was
eliminated from the NA gene of the vaccine virus used for production of clinical lot material.1691 Of note,
candidate vaccine viruses (CVVs) are not typically tested for antiviral sensitivity as part of the routine set
of characterization assays performed prior to release of CVVs to manufacturers.1692,1693,1694
1685 Baz M et al (2010) Effect of the neuraminidase mutation H274Y conferring resistance to oseltamivir on the replicative
capacity and virulence of old and recent human influenza A(H1N1) viruses. J Infect Dis 201: 740-745
1686 Kaminski MM et al (2013) Pandemic 2009 H1N1 influenza A virus carrying a Q136K mutation in the neuraminidase gene
is resistant to zanamivir but exhibits reduced fitness in the guinea pig transmission model. Journal of virology 87: 1912-
1915
1692 Vaccine response to the avian influenza A(H7N9) outbreak
- step 1: development and distribution of candidate vaccine viruses.
http://www.who.int/influenza/vaccines/virus/CandidateVaccineVirusesH7N9_02May13.pdf. Last Update Accessed
September 14, 2015.
1693 Update of WHO biosafety risk assessment and guidelines for the production and quality control of human
influenza vaccines against avian influenza A(H7N9) virus.
http://www.who.int/biologicals/areas/vaccines/influenza/biosafety_risk_assessment_10may2013.pdf. Last Update Accessed
September 14, 2015.

The strengths and limitations of GoF and alt-GoF approaches for identifying antiviral resistance markers
that can be removed from vaccine viruses are summarized in Table 15.35. Both GoF and alt-GoF
approaches can be used to identify mutations that confer antiviral resistance to currently circulating
influenza strains. GoF approaches are relatively more efficient and effective for the discovery of new
antiviral resistance markers but may uncover mutations that do not yet exist in nature, which is not
relevant for this application because vaccine viruses are based on wild type viruses. The FDA, which
must approve of all vaccine strains that are used for large-scale production, prefers that the HA and NA
genes of a vaccine strain are as close to the wild type strain as possible.1695 As a result, markers without
causal links to antiviral resistance across multiple strain contexts may not be approved for this application
(though approval would depend on the level of manipulation and would be considered on a case-by-case
basis). Both GoF and alt-GoF approaches can provide this information. GoF approaches, namely targeted
mutagenesis of antiviral-sensitive strains to introduce mutations expected to confer antiviral resistance,
can be used to demonstrate that a particular mutation (or set of mutations) is necessary and sufficient to
confer resistance. Alternative approaches, namely targeted mutagenesis of antiviral-resistant strains to
introduce mutations expected to restore antiviral sensitivity, can be used to demonstrate that a particular
amino acid or set of amino acids are necessary for antiviral resistance. As the ultimate goal is to restore
antiviral sensitivity to vaccine strains harboring NAI-resistant NA genes, either approach is equally
suitable for identifying molecular markers linked to antiviral resistance for this purpose.
Benefits to vaccine development: Targeted Mutagenesis to Remove Antiviral Resistance Markers from
Vaccine Viruses
• Efficient and effective approaches for
GoF Experimental Approaches: discovering new mutations that confer
• Serial passaging of viruses in the antiviral resistance to currently • May uncover antiviral
presence of therapeutic circulating strains resistance mutations that
• Targeted GoF can be used to do not yet exist in nature,
• Genetic modification of antiviral-
demonstrate that particular mutations which are not relevant
sensitive strains to introduce
are necessary and sufficient to confer for this application
genetic traits expected to confer
antiviral resistance antiviral resistance across multiple
strain contexts
Alt-GoF Experimental • Discover new mutations that confer
Approaches:
• Approaches are less
antiviral resistance to currently
efficient and effective for
• Genetic modification of antiviral- circulating strains
the discovery of novel
resistant strains to introduce traits • Targeted LoF can be used to antiviral resistance
expected to restore antiviral demonstrate that particular mutations markers than GoF
sensitivity are necessary for antiviral resistance approaches
• Other approaches (see table 15.32) across multiple strain contexts
15.7.7 GoF Benefits to the Development of Therapeutics
Only two classes of FDA-approved antivirals are approved for use in the U.S.: the adamantanes, which
inhibit the viral M2 protein, and the neuramidase inhibitors (NAIs), which include zanamivir (Relenza),
1695 Ibid.

oseltamivir (Tamiflu), and peramivir (Rapivab).1696 The adamantanes are no longer recommended for use
due to widespread resistance. Single mutations are sufficient to confer resistance to one or multiple NAIs
and have been observed in nature, though NAI-resistance mutations are not yet widespread. Moreover,
the NAIs exhibit limited efficacy, especially if administered more than 48 hours after symptom onset.1697
Thus, there is an urgent need for the development of new therapeutics against influenza viruses.1698
15.7.7.1 Therapeutic Development Benefit 1: Inform Development of Therapeutic Candidates
Given the high mutation rate of influenza viruses, viruses can readily acquire mutations to many
therapeutics. Screening potential therapeutics based on how readily antiviral resistance emerges provides
one mechanism for differentiating between therapeutic candidates based on their likely field efficacy.
Prior to field deployment of a therapeutic, serially passaging viruses in the presence of therapeutic, a GoF
approach, is uniquely capable of determining whether and how readily resistance arises. Furthermore, as
resistance is expected to arise in human populations following deployment of the therapeutic, determining
whether resistance enhances the infectivity, transmissibility, or virulence of a virus is an important aspect
of safety testing of the therapeutic candidate. Taken together, GoF approaches are critical for testing the
potential efficacy and safety of new therapeutic candidates.
15.7.7.2 Therapeutic Development Benefit 2: Facilitate Regulatory Approval of Therapeutic

Candidates
investigational product, and the genotypic and phenotypic characterization of resistant viruses.1699
Mechanism of action studies should demonstrate the investigational product’s ability to specifically
inhibit viral replication or virus-specific function and should establish the site of the product’s action.
GoF approaches have the potential to support two aspects of an IND application for therapeutics in
development: (1) determination of the mechanism of action of a therapeutic, and (2) the in vitro selection
of resistant viruses.
15.7.7.2.1 Determining the Mechanism of Action of a Therapeutic
the licensing process.1700 The influenza field is pursuing multiple strategies for developing new
therapeutic candidates, including the deliberate design or selection of therapeutics targeting specific viral
or host proteins and high-throughput screening of libraries of small molecules to identify compounds that
reduce viral replication in vitro. In the former case, the drug target of the therapeutic candidate is known,

1697 CDC. Use of Antivirals. Background and Guidance on the Use of Influenza Antiviral Agents.
http://www.cdc.gov/flu/professionals/antivirals/antiviral-use-influenza.htm. Last Update February 25, 2015. Accessed
November 28, 2015.
1700 Ibid.

while in the latter case, the therapeutic target is unknown. GoF approaches can be used to gain insight into
the mechanism of activity of therapeutics that directly target virus proteins, thus benefitting the
development of new drugs. Below, the benefits of GoF approaches, relative to alternative experimental
approaches, for the determination of antiviral mechanisms in both of these scenarios is evaluated.
Benefits and Limitations of GoF Approaches
by the therapeutic, or the phenotypic consequences of mutations may be unclear.1701,1702,1703 Additionally,
if the drug targets a host protein, this approach provides indirect information about its mechanism of
activity, which must be inferred based on prior knowledge of virus-host interactions.
Benefits and Limitations of Alt-GoF Approaches
information about potential viral targets of therapeutics. Viral targets must be inferred based on prior
knowledge of virus-host interactions, which is likely to be challenging. Furthermore, because this kind of
indirect information does not provide insight into antiviral mechanisms, this host-focused approach is of
limited value for the study of therapeutics with known viral targets.
642-646

a process in the viral life cycle may be difficult from such a static snapshot. In addition, this approach is
less suitable for investigating therapeutics that target a protein-protein or protein-nucleic acid complex
(either a virus-host complex or a virus-virus complex), either to inhibit the function or block the
formation of the complex. The relevant interaction partner may be unknown, or recombinantly producing
and crystallizing the protein complex may be difficult. Critically, because of the high level of effort
required for X-ray crystallography, it is not a feasible approach for simply screening the potential viral
targets of an unknown antiviral.
of the location of the drug binding site.1707 However, inferring the mechanism of antiviral activity based
on knowledge about the drug-virus protein interaction may be difficult, and the approach is less suitable
for studying therapeutics that target a protein-protein or protein-nucleic acid complex (either a virus-host
complex or a virus-virus complex).
Summary – Benefits of GoF Approaches Relative to Alt-GoF Approaches
The strengths and limitations of GoF and alt-GoF approaches that can provide insight into the mechanism
of action of a candidate therapeutic are summarized in Table 15.36. Taken together, serial passaging of a
virus in the presence of therapeutic to discover mutations that confer resistance, a GoF approach, is
uniquely capable of identifying the viral target of a novel therapeutic with an unknown mechanism of
action. Given that researchers are undertaking unbiased screens to identify candidate therapeutics that
inhibit viral replication, this represents a valuable benefit for the development of new influenza
therapeutics. For therapeutics with known viral targets, this information about resistance mutations can
provide foundational information to guide follow-up structural, cell biological, and biochemical studies
investigating the mechanism of action of the therapeutic. Although crystallography and photoaffinity
cross-linking can also provide insight into the antiviral mechanisms of therapeutics that directly bind to
and inhibit virus proteins, inferring mechanistic information based on static information about the virus-
antiviral complex may be difficult. In these cases, knowledge about mutations that confer resistance,
generated through GoF approaches, provides an additional source of information that can be used to
generate testable hypotheses about mechanism of antiviral activity. Finally, the identification of host
factors that are required for antiviral activity is a critical aspect of examining therapeutics with unknown
targets. Though solely using host-focused approaches to elucidate the antiviral mechanism of a
therapeutic that targets the virus would be difficult, this information complements GoF approaches to
strengthen the evidence base for the drug’s mechanism of action.

Benefits to therapeutic development: Identify the mechanism of action of a candidate therapeutic
• Identify the viral protein target of a candidate • Elucidating the mechanism of action of a therapeutic based
on indirect information about resistance mutations may be
therapeutic with an unknown target
GoF #1: difficult
Serial passaging of viruses in the
• Provide insight into the mechanism of action o Resistance mutations may arise in non-target proteins,
of the therapeutic through the identification of confounding interpretation of results
presence of therapeutic
mutations that confer resistance
• Not suitable for identifying the targets of therapeutics that
target host proteins
Alt-GoF #1:
RNAi screen targeting host proteins to • Identify the host protein target of a candidate
therapeutic with an unknown target • Provides indirect information about the viral protein targets
identify host proteins that are critical
of a therapeutic
for the antiviral activity of a
therapeutic

Alt-GoF #2: • Provides direct information about the
interaction between a therapeutic and its viral • Inferring mechanism of activity based on static information
Analyze the crystal structure of a protein target about the therapeutic-viral protein interaction may be
therapeutic in complex with its viral difficult
protein target o May provide insight into the mechanism of

• Provides direct information about the binding
Alt-GoF #3: site of a therapeutic on its viral protein target • Inferring mechanism of activity based on static information
about the therapeutic binding site may be difficult
Photo-affinity crosslinking • May provide insight into the mechanism of

15.7.7.2.2 Determining the Genetic Threshold for Resistance Development
resistant viruses and phenotypic characterization of resistant viruses.1708 Selection for resistance studies
should be repeated multiple times to determine if the same or different patterns of resistance mutations
develop, as well as to determine how the concentration of the therapeutic impacts how readily resistance
develops. These studies constitute GoF approaches. The FDA guidance does not suggest any alternative
approaches that could provide similar information. In fact, prior to deployment of the therapeutic and the
emergence of resistant viruses in nature, no alternative approaches can provide this information. Thus,
essential for the licensing of new therapeutics.
15.7.7.3 Therapeutic Development Benefit 3: Inform Guidelines for Use of Therapeutics
whether and how readily antiviral resistance arises. Given that influenza viruses readily acquire resistance
to NAIs (i.e. upon acquisition of one or two mutations), influenza researchers cited a lack of knowledge
about the potential utility of combination therapies as a critical gap in public health preparedness for
influenza epidemics and pandemics.1709 In addition, understanding whether antiviral resistance arises
more readily or differently in at-risk populations, such as obese or immunocompromised people, in either
scenario can provide information that further refines therapeutic guidelines. GoF approaches can address
each of these questions.
administration of multiple therapeutics in combination. First, serial passaging of virus in animals dosed
with varying amounts of the therapeutic provides insight into the dose-dependence of the emergence of
resistant viruses. Because host-dependent factors, such as the rate of metabolism or clearance of the
therapeutic, influence the concentration of therapeutic the virus experiences, conducting passaging studies
in animals provides more relevant information than in vitro passaging studies. Second, serial passaging of
virus in cells or in animals in the presence of multiple mAbs (or other types of therapeutics) can be used
to determine how readily resistance arises in response to combination versus single therapies. Although in
vitro selection studies are useful for screening different combinations of therapeutics, because of the role
of bioavailability and other host-dependent factors on antiviral efficacy, all promising combination
therapies should be validated through in vivo passaging experiments. In both cases, serial passaging of
virus in mouse models for at-risk populations (e.g. immunocompromised mice or obese mice) provides
additional information about the extent to which the likelihood of resistance or patterns of resistance
mutations vary depending on host factors, which may inform therapeutic guidelines for specific at-risk
populations. No alternative approaches are capable of providing similar information about the dose-
dependence of resistance or whether combination therapies lead to resistance less readily than individual
therapies.

Taken together, GoF approaches that lead to the generation of viruses that are resistant to therapeutics in
development are uniquely capable of determining the therapeutic dose that is least likely to lead to the
acquisition of antiviral resistance as well as determining whether combination therapies better prevent the
emergence of resistant viruses than individual therapies. Both types of information benefit the
development of therapeutic strategies that will be effective for a longer period of time in the field.
15.8 Influenza viruses: Detailed analysis of the benefits of GoF research involving
reassortment
This assessment describes the benefits of GoF experimental approaches that aim to assess the genetic
compatibility and fitness of viruses following reassortment. While the phenotypic consequences of
reassortment events between two viruses cannot be predicted with certainty, reassortant strains may
exhibit enhanced fitness, pathogenicity, and/or transmissibility relative to one or both parental strains.
(Notably, reassortant strains may also display reduced fitness, pathogenicity, and/or transmissibility
relative to parental viruses.) In this section, we provide an overview of GoF approaches that can be used
to assess the reassortment potential between two viruses and describe the scientific outcomes of each
approach. Each approach will be discussed in more detail in the context of our detailed analysis of the
benefits of GoF and alt-GoF research that can provide insight into the genetic compatibility and
reassortment potential of multiple viruses, below.
15.8.1.1 Targeted reassortment by combining viral gene segments from two or more viruses to generate
viable reassortant viruses
Targeted reassortment of virus gene segments from two or more wild type virus isolates followed by
characterization of fitness in cell culture or in representative animal models is used to assess genetic
compatibility. This approach is in part performed to evaluate the genetic compatibility and viability of a
single reassortant virus, which can inform the understanding of the mechanisms underlying genetic
compatibility between virus gene segments across virus strains and subtypes. For example, a reassortant
virus compromised of virus gene segments sharing homology to the 1918 H1N1 pandemic virus from
eight different wild type avian isolates was generated to demonstrate that some 1918-like avian viruses
circulating in nature (which reassort frequently) are genetically compatible.1710
15.8.1.2 Forward genetic screen to identify viable reassortant viruses
Forward genetic screens involve the generation of a panel of clonal recombinant viruses by
comprehensive reassortment of parental gene segments from two viruses (i.e. all or many possible gene
combinations), followed by characterization of the fitness of reassortants in appropriate mammalian
model systems. Follow-up studies may be performed to evaluate pathogenicity, infectivity, and/or
transmissibility of viable reassortants. This approach is performed to evaluate viability and genetic
compatibility of reassortant viruses, which provides a foundation for studies investigating mechanisms
governing reassortment and informs the potential for reassortant viruses to emerge in nature and the
potential public health consequences of such an emergence event.
1710 Watanabe T et al (2014) Circulating avian influenza viruses closely related to the 1918 virus have pandemic potential. Cell

15.8.1.3 Non-targeted reassortment using reverse genetics to select for viable reassortant viruses
In this approach, reassortants are generated using reverse genetics to mix viral gene segments of two wild
type viruses (i.e. mix up to 16 gene segments in total) in the context of cell culture or animal models. Use
of cell culture model systems involves the transient transfection of viral gene segments, while the in vivo
method involves the inoculation of ferrets with transiently transfected cells, followed by viral
reassortment in vivo. Both approaches are followed by limited passaging to select for viable reassortants.
Clonal isolates that emerge are then genotyped to identify their gene composition. This approach provides
insight into viable gene reassortment combinations as well as the relative fitness of reassortants under
selection pressures, which informs the potential and likelihood of reassortment emergence in nature.
15.8.1.4 Co-infection to select for viable reassortant viruses
In this approach, cultured cells or representative animal models are co-infected with two different wild
type viruses, followed by genotyping of clonal isolates that emerge during co-infection. This approach
determines the viability of various gene reassortment combinations and the relative fitness of reassortants
under selection pressures, which can inform the potential and likelihood of emergence in nature.
15.8.2 Overview of the potential benefits of GoF experiments involving reassortment
Here we evaluate whether any of the GoF Influenza approaches have the potential to benefit each of the
Assessments of Gain-of-Function Research.” We also describe additional benefit areas we identified
during our research. Each potential benefit will be analyzed in detail below.
GoF approaches benefit scientific knowledge by providing insight into the reassortment potential between
different virus strains, including human seasonal and animal strains, two different human seasonal strains,
two different animal strain sub-types, and two different animal strains within the same sub-type.
Specifically, GoF approaches can determine the genetic compatibility between virus strains and the
phenotypic properties of reassortant viruses, including fitness, transmissibility, and virulence.
GoF approaches may benefit surveillance for reassortant viruses. Specifically, information about the
phenotypic properties of reassortant viruses may inform assessment of the risks posed by reassortant
viruses detected in nature.
15.8.2.3 Vaccines, therapeutics, and diagnostics
GoF-derived information about the reassortment potential of two different viruses is not relevant for the
development of vaccines or therapeutics.
As existing influenza diagnostics are not equipped to rapidly screen and detect reassortants, information
about reassortants with phenotypic properties of concern could, in principle, guide development of
diagnostics to detect those reassortants. However, GoF approaches do not provide insight into the
likelihood that reassortment will occur in nature, which is a function of complex ecological factors that
govern the likelihood of co-infections. The likelihood of reassortment is also a critical factor for the
design of targeted diagnostics for reassortant viruses (i.e. there is no need to design diagnostics for rare

reassortant events). For this reason, GoF approaches are unlikely to trigger the development of new
diagnostics independently of the observation of co-infection or reassortment events occurring in nature.
the results of reassortment studies may stimulate risk mitigation activities to limit the potential for “risky”
pandemic vaccine development and other pandemic preparedness initiatives.
No economic benefits of GoF reassortment studies were identified.
Here, the ability of GoF approaches to address a key outstanding question related to the reassortment of
influenza viruses in humans and other host species is evaluated:
• What is the potential for reassortment between two influenza virus strains?
o Are two influenza viruses genetically compatible?
o What is the relative fitness of reassortants that may affect the likelihood of their emergence
under selection in a host?
o How do selection pressures influence reassortment?
Reassortment involves the exchange of one or more complete virus gene segments between two different
viruses during the co-infection of a single cell. The process of reassortment contributes to influenza virus
evolution and viral diversity by allowing the rapid exchange of genetic and phenotypic properties under
selection pressure and can result in viruses that display enhanced fitness, immune evasion and antigen
escape, and resistance to antivirals.1711 Notable examples of the role of reassortment in influenza virus
biology include the reassortment of seasonal and animal influenza viruses leading to the emergence of
human pandemic viruses with minimal population immunity.1712 Considerable gaps in knowledge remain
about the biology and prevalence of reassortment in nature within and across host populations.
Accordingly, whether such events will occur and will lead to the generation of viruses with enhanced
fitness, pathogenicity, and/or transmissibility is not understood. Although many of the unknowns
regarding reassortment fall outside the scope of GoF research, GoF approaches can be used to understand
whether two viruses are genetically compatible, which provides a foundation for follow-up studies
investigating the mechanistic basis of genetic compatibility. These studies include efforts to identify how
multiple viral gene segments cooperate to shape other viral phenotypes such as replicative fitness and
efficient cell entry and exit. (We note that follow-up studies that are focused on characterizing the
pathogenicity or transmissibility of reassortant viruses are covered separately, in the relevant GoF
phenotype sections, 15.3 and 15.4.)
1712 Scholtissek C (1994) Source for influenza pandemics. European journal of epidemiology 10: 455-458

15.8.3.1 Scientific Knowledge Gap: What is the potential/capability for reassortment between two
influenza virus strains?
Several GoF approaches can lead to the generation of reassortant viruses:
• Targeted reassortment to generate a virus comprised of gene segments from two or more wild
type isolates,
• Forward genetic screens involving comprehensive reassortment to generate a panel of clonal viral
isolates followed by assessment of fitness in cell culture or representative animal models,
• Non-targeted reassortment involving gene segments from two different viruses to generate a
mixed population of reassortant viruses, followed by selection for compatible virus genotypes in
cell culture or representative animal models, and
• Co-infection of cell culture or representative animal models with two different viruses to select
for compatible virus genotypes.
Collectively, these approaches definitively demonstrate whether reassortment can occur between wild
type viruses and enable the identification of reassortment gene combinations that permit replication in in
vitro or in vivo model systems. This provides insight into the genetic compatibility of virus gene
segments. For the targeted reassortment approach, viral gene segments are selected based on a property of
interest (e.g. homology to a human pandemic virus) to answer a specific question about the genetic
compatibility between two or more viruses, which differs from the other GoF approaches that more
broadly query the range of reassortment combinations that are possible between two viruses. Because
forward genetic screens individually test every possible gene combination between two viruses, this GoF
approach can assess the viability and fitness of each viral clone that may otherwise be missed with
selection based approaches (below) in which more fit clones outcompete. However, the outcomes
associated with forward genetic screens are independent of the selection pressures that shape reassortment
potential and viral population diversity, and therefore may not fully represent the likelihood of
reassortants emerging. The use of non-targeted reassortment by transfection of cell culture model systems
with gene segments from two separate viruses to select and identify emergent reassortants presents
several different advantages. First, this approach provides insight into how host pressures and competition
among reassortants shapes outcomes. Importantly, this approach can evaluate selection pressures
independent from the requirement of co-infection of the same host cell, and is therefore not impacted by
differences in the receptor specificity and efficiency of cell entry of parental viruses. Second, the ability to
selectively remove a single gene segment that may otherwise outcompete or skew virus populations
enables assessment of the compatibility of many gene segment combinations, relative to the co-infection
approach. Similar to the non-targeted reassortment approach, the co-infection approach provides insight
into how the host pressure and competition impact selection. A major benefit of this approach is that it
mimics the natural scenario under which reassortment can occur. However, in the event that two viruses
of interest display different tissue and cell tropism or significant disparity in fitness or infectivity, this
approach permits study of a limited number of reassortment combinations. For all three approaches, the
use of in vivo animal models for reassortment studies provides more relevant information due to the
complexity of host selection pressures relative to cell culture models. All GoF approaches described here
depend on whether the mechanisms and selection pressures underlying fitness and reassortment in cell
culture or animal models are representative of those in humans and whether the genetic compatibility
observed for the select number of strains analyzed is generalizable in other virus contexts. Moreover, the

use of the methods above may not capture the dynamics of co-infection and reassortment in nature, which
are likely dependent on the time and exposure to influenza viruses in addition to the factors discussed
above like disparities in fitness among viral isolates in humans.
A select number of alt-GoF approaches can be used to analyze the reassortment potential of two different
viruses. Analyzing the sequences of human and animal surveillance isolates to detect reassortment events
can provide insight into the occurrence and prevalence of reassortment in nature. This approach includes
sequence inspection for several different types of reassortment events, involving:
• Two different human seasonal virus sub-types (e.g. H1N1 and H3N2),
• Human or animal virus strains within the same sub-type (e.g. different clades of H3N2),
• Human and animal viruses (e.g. human seasonal H3N2 and swine-origin H1N1), and
• Two different animal virus sub-types (e.g. H9N2 and H7Nx).
Analysis of both animal and human isolates provides information that is applicable to a broad number of
strains, and the analysis of human isolates provides information about reassortment potential that is
directly relevant to human populations. However, this approach is significantly limited by the quality and
availability of existing genetic surveillance data. Reassortment events are most commonly identified
through individual phylogenetic analysis of each viral gene segment to identify origin and ancestry. 1713
This requires full genome sequences and large sequence databases for effective determination of
phylogenetic ancestry, which are not always available. Furthermore, in cases of low genetic diversity
between parental strains, distinguishing between mutations and reassortment events may be difficult,
while in cases of high viral diversity between parental strains, distinguishing between reassortant and wild
type gene segments may be difficult. In addition, this analysis is limited to the study of reassortant viruses
that have evolved (and have been subsequently detected) in nature.
A second type of alt-GoF approach involves the analysis of viral isolates from humans or animals that
have been co-infected with two influenza viruses. This approach can determine whether reassortment has
occurred and also may provide insight into the genetic compatibility of various gene combinations, as
well as host selection pressures that shape the outcome of reassortment events. That analysis of human
and animal isolates provides information that is directly relevant to reassortment potential in nature is a
strength of this method. However, this approach is also subject to significant limitations. Although co-
infection events occur, the success of this approach depends on the occurrence of productive co-infection
and the collection of samples for later analysis. Because the frequency and distribution of co-infection
across host species and populations is unknown, designing systematic sampling strategies for detecting
co-infection events would be difficult. Rather, these events are captured on an ad hoc basis. Moreover,
unknowns in the route of infection, the level and time of exposure, and diversity in the host response due
to existing natural or induced immunity limits the ability of this approach to reliably assess genetic
compatibility of reassortant viruses. Similarly, if the viruses analyzed have disparate tissue and cell
tropism or fitness in vivo, this approach may not accurately portray reassortment potential.
The use of replication incompetent viruses provides another alternative method for the analysis of genetic
compatibility between gene segments from two influenza viruses. In these model systems, viral
replication can be assessed in cell culture lines that are engineered to stably express an essential viral
protein that is missing from the “replication-incompetent” virus strains used for infection. For example,
the replacement of the PB2 gene with a GFP-expression construct that has the necessary flanking, non-
coding and packaging sequences from the viral genome can only replicate in cell lines that stably express

exogenous PB2.1714 The result is a virus that is biologically constrained to replication in that cell line.
Several replication incompetent model systems have been made, and these systems have been used to
assess the of genetic compatibility of virus gene segments by targeted reassortment resulting in the
generation of a clonal replication incompetent virus.1715,1716,1717 For example, the genetic background of a
lab-adapted strain was compatible with the HA and NA of a high pathogenicity avian H5N1 virus.1718
However, this system has not yet been used to broadly assess reassortment potential by the identification
of replication incompetent reassortant viruses from a mixed population after transfection of 16 gene
segments or fewer, as is the case during co-infection. One major drawback is that this approach does not
capture the complex selection pressures observed in vivo. Additionally, results may not translate to
reassortment in humans, and findings may not be generalizable to other virus contexts.
A final alt-GoF approach utilizes in vitro virus-free methods to investigate genetic compatibility of viral
gene segments in isolation. In particular, forward genetic screens can be used to identify novel gene
segment combinations or reassortment events that contribute to a phenotype underlying viral fitness and
infectivity, such as polymerase activity. Though the simplicity and relatively high-throughput nature of
these methods renders them appealing as a screening approach for the evaluation of genetic compatibility
between two viruses, these approaches are inherently limited to the characterization of phenotypes
previously identified in other experiments. In addition, results may not be recapitulated in the context of
the full virus or in vivo.
Table 15.37 summarizes the benefits and limitations of GoF and alt-GoF approaches that assess the
potential for reassortment between two wild type viruses. Taken together, GoF approaches are uniquely
capable of proactively assessing the potential for any two influenza viruses to reassort, as well as for
comprehensively evaluating the viability of various gene combinations. Notably, the outcomes of forced
laboratory reassortment events may provide limited insight into the likelihood that such reassortment
events will occur in nature, as natural reassortment depends on complex factors such as the rate of co-
infection and the distribution of genetically compatible viruses (which are unknown). In addition, the
relevance of this information for human populations depends on the suitability of animal models.
Although surveillance-based approaches can provide broad insight into of the prevalence and distribution
of reassortment viruses in different host populations, their utility is severely limited by the quality and
availability of surveillance data. Similarly, the analysis of humans or animal isolates during co-infection
is an unreliable method for determining the reassortment potential and genetic compatibility of two
viruses, and opportunities for such studies are rare. The use of replication incompetent viruses is a
promising approach for assessment of the genetic compatibility and reassortment potential between two
viruses, but this system is not commonly used for this purpose and requires further validation. Moreover,
it cannot capture the complex selection pressures observed in vivo and may not translate to mechanisms of
reassortment in humans. Although the use of in vitro virus free systems is useful from an initial screening
approach, results may not be recapitulated during the complete viral life cycle.
1715 Ibid.
the Detection of Influenza Virus Neutralizing Antibodies. J Virol 84: 2157-2163
1717 Baker SF et al (2014) Influenza A and B virus intertypic reassortment through compatible viral packaging signals. Journal
of virology 88: 10778-10791

Table 15.37. Summary of the Benefits of GoF Approaches For the Study of Reassortment
Scientific Knowledge Benefits— What is the potential for reassortment between two influenza virus strains?
• Narrow breadth: Results may not generalize to other
virus strains
• Translatability: Results from representative animal
models may not translate to humans
• Artificiality: Laboratory results may not reflect
GoF #1a [1]a: likelihood of reassortment or behavior of reassortant
Targeted reassortment by combining viral gene viruses in nature
segments from two or more virus genotypes to o Likelihood of multiple reassortant events is lower
generate a single virus (in vitro) than with reassortment between two viruses
• Determine whether two viruses can o Approach does not capture how selection pressures
reassort and relative fitness among virus genotypes influence
reassortment outcomes
o Can assess viability and compatibility of
select/specific gene segment o Approach does not capture impact of time and dose
combinations, independent of relative on reassortment outcomes
fitness (of wild type versus reassortant • Narrow breadth: Results may not generalize to other
viruses) virus strains
• Proactive – can be performed using virus • Translatability: Results from representative animal
gene segments that have not reassorted in models may not translate to humans
nature
• Artificiality: Laboratory results may not reflect
GoF #1b [1]: likelihood of reassortment or behavior of reassortant
Targeted reassortment by combining viral gene viruses in nature
segments from two or more virus genotypes to o Approach does not capture how selection pressures
generate a single virus (in vivo) and relative fitness among virus genotypes influence
o Approach does not capture impact of time and dose
on reassortment outcomes
• Selection pressures in vitro are less complex than in
vivo

• Narrow breadth – Results may not generalize to other
virus strains
• Translatability – Results from representative animal
models may not translate to humans
GoF #2a [2]:
Forward genetic screen by comprehensive • Artificiality – Laboratory results may not reflect
targeted reassortment generating a panel of likelihood of reassortment or behavior of reassortant
clonal reassortants to identify compatible virus viruses in nature
genotypes (in vivo) o Approach does not capture how selection pressures
and relative fitness among virus genotypes influence
• Determine whether two viruses can reassortment outcomes
reassort o Approach does not capture impact of time and dose
o Can assess viability and compatibility of on reassortment outcomes
all possible gene segment combinations,
independent of relative fitness (of wild • Narrow breadth – Results may not generalize to other
type versus reassortant viruses) virus strains
• Proactive – can be performed using virus • Translatability – Results from cell culture models may
gene segments that have not reassorted in not translate to humans
nature • Artificiality – Laboratory results may not reflect
GoF #2b [2]:
likelihood of reassortment or behavior of reassortant
Forward genetic screen by comprehensive viruses in nature
targeted reassortment generating a panel of
clonal reassortants to identify compatible virus o Approach does not capture how selection pressures
genotypes (in vitro) and relative fitness among virus genotypes influence
vivo

virus strains
GoF #3a [3]:
• Translatability – Results from representative animal
Non-targeted reassortment with up to 16 gene • Determine whether two viruses can models may not translate to humans
segments from two different viruses generating reassort
a mixed population of recombinant viruses to • Artificiality – Laboratory results may not reflect
select for compatible virus genotypes with o Can assess viability and compatibility of likelihood of reassortment or behavior of reassortant
enhanced infectivity (in vivo) many gene segment combinations by viruses in nature
controlling for disparity in fitness o Approach does not capture impact of time and dose
between reassortants and wild type on reassortment outcomes
viruses
• Proactive – can be performed using virus • Narrow breadth – Results may not generalize to other
gene segments that have not reassorted in virus strains
nature • Translatability – Results from cell culture models may
GoF #3b [3]: not translate to humans
• Gain insight into how host pressures
Non-targeted reassortment with up to 16 gene
segments from two different viruses generating
influence reassortment outcomes and • Artificiality – Laboratory results may not reflect
population frequency likelihood of reassortment or behavior of reassortant
a mixed population of recombinant viruses to
o Can evaluate selection pressures viruses in nature
select for compatible virus genotypes with
enhanced fitness (in vitro) independent from co-infection o Approach does not capture impact of time and dose
vivo

virus strains
• Determine whether two viruses can • Translatability – Results from representative animal
reassort models may not translate to humans
o Gain insight into the compatibility of • Artificiality – Laboratory results may not reflect
virus gene segments likelihood of reassortment or behavior of reassortant
GoF #4a [4]: viruses in nature
• Proactive – can be performed using viruses
Co-infection with two wild-type viruses to o Approach may not capture impact of time and dose
that have not reassorted in nature
select for compatible virus genotypes with on reassortment outcomes
enhanced infectivity (in vivo) • Gain insight into how host pressures
influence reassortment outcomes and o Approach may not capture the full potential for
population frequency reassortment if there are large disparities in fitness
between wild type viruses and other reassortants or if
o Mimics natural scenario under which there is inefficient co-infection in vivo
reassortment can occur
o Viruses displaying distinct tissue/cell tropism in
representative model systems are less likely to
reassort

• Determine whether two viruses can virus strains
reassort • Translatability – Results from cell culture models may
o Gain insight into the compatibility of not translate to humans
virus gene segments
• Artificiality – Laboratory results may not reflect
GoF #4b [4]: • Proactive – can be performed using viruses likelihood of reassortment or behavior of reassortant
Co-infection with two wild-type viruses to that have not reassorted in nature viruses in nature
select for compatible virus genotypes with • Gain insight into how host pressures o Approach does not capture impact of time and dose
enhanced fitness (in vitro) influence reassortment outcomes and on reassortment outcomes
population frequency o Approach may not capture the full potential for
o Can evaluate selection pressures reassortment if there are large disparities in fitness
independent from co-infection (as this between wild type viruses and other reassortants
can be controlled for in vitro) • Selection pressures in vitro are less complex than in
vivo

• Reactive – Involves analysis of viral isolates that
already exist in nature
• Translatability – Results may not translate to
reassortment in humans when animal isolates are
• Determine whether reassortant viruses exist analyzed
in nature o Whether animals under study are representative
o Directly translates to human disease models for human disease has not been established
Alt-GoF #1 [1]: when human isolates are analyzed when • Limited by the quality and availability of surveillance
Analysis of surveillance data to determine the applicable data
occurrence and prevalence of reassortment o Analyzes broad data sets applicable to o Incomplete genome sequences limit the identification
many strains of gene segment ancestry (i.e. reassortants)
o Gain insight into the prevalence and o Reassortment between genetically similar strains may
distribution of reassortant viruses across not be evident or distinguishable from genetic drift
host populations
o Requires large data sets for reliable phylogenetic
analysis
o High viral diversity, as observed in avian populations,
limits the ability to distinguish between reassortment
and wild type gene segments

• Reactive – Analysis of viral isolates that already exist in
nature
• Determine whether reassortant viruses exist • Reassortment between genetically similar strains may
in nature not be evident or distinguishable from genetic drift
Alt-GoF #2 [2]:
o Directly translates to human disease • Likelihood of sample collection – Co-infection events
Analysis of viral isolates from humans that may be rare; Viruses displaying distinct tissue/cell
have been co-infected with two influenza o Gain insight into the genetic
compatibility of virus gene segments tropism are less likely to reassort; Isolates from patients
viruses that are infected with two viruses may not be collected,
• Gain insight into how host pressures identified, or saved for further analysis
influence reassortment outcomes and
population frequency • Human populations display variable immune responses
due to differences in vaccination, previous exposures to
influenza, and host factors complicating interpretation
of selection pressures impacting reassortment
• Reactive – Analysis of viral isolates that already exist in

nature
• Reassortment between genetically similar strains may
• Determine whether reassortant viruses exist not be evident or distinguishable from genetic drift
in nature • Likelihood of sample collection – Co-infection events
Alt-GoF #3 [3]:
o Gain insight into the genetic may be rare; Viruses displaying distinct tissue/cell
Analysis of viral isolates from animals that compatibility of virus gene segments tropism are less likely to reassort; Isolates from animals
have been co-infected with two influenza
• Gain insight into how host pressures that are infected with two viruses may not be collected,
viruses
influence reassortment outcomes and identified, or saved for further analysis
population frequency • Translatability – Results may not translate to
reassortment in humans
o Whether animals under study are representative
models for human disease has not been established

• Translatability – Results may not translate to
reassortment in humans
virus strains
• Limited Utility – Replication incompetent systems have
only been developed and validated for a limited number
of strains
Alt-GoF #4 [4]: o Use of existing models that make use of gene
In vitro, replication incompetent model system:
• Gain insight into genetic compatibility of
virus gene segments segments derived from lab-adapted strains will
Targeted reassortment by combining viral gene depend on genetic compatibility
segments from two or more virus genotypes to • Proactive – can be performed using virus
gene segments that have not reassorted in • Selection pressures in vitro are less complex than in
generate a single virusb vivo
nature
• Artificiality – Laboratory results may not reflect
likelihood of reassortment or behavior of reassortant
viruses in nature
o Approach does not capture how selection pressures
and relative fitness among virus genotypes influence
Alt-GoF #5 [5]:
• Gain insight into the compatibility of virus • Simplicity of model system – Results based on the
gene segments study of a viral protein/phenotype in isolation may not
In vitro, virus-free: Forward genetic screen to be recapitulated in the context of the full virus
evaluate genetic compatibility of virus gene • Proactive – can be performed using virus
segments for a phenotype underlying fitness gene segments that have not reassorted in • Narrow breadth – Results may not generalize to other
nature virus strains
a
b
To date, replication incompetent systems have only been used for targeted reassortment experiments, but in principle these systems could be used for non-
targeted reassortment studies (i.e. transfection of cells with multiple gene segments from two or more viruses to broadly assess reassortment compatibility.)

15.8.4 Benefits of GoF approaches to Surveillance
The importance of reassortment in influenza virus biology is highlighted by its role in the emergence of
human pandemic viruses with minimal population immunity – all four of the influenza pandemics that
have occurred over the past century were likely caused by strains that arose through reassortment between
influenza viruses, although the role of reassortment in the emergence of the 1918 pandemic virus is
controversial.1719,1720,1721,1722,1723 While the emergence of reassortant viruses cannot yet be predicted,
surveillance for reassortant viruses to assess their occurrence and prevalence in nature is of interest for
pandemic preparedness, and as such is one of the factors considered in pandemic risk assessments
(discussed further below). Given the importance of epistasis in influenza biology, determining whether a
reassortant virus poses an increased risk to human populations relative to its parental viruses poses a
major challenge.
GoF approaches that proactively determine the reassortment potential between two viruses and
phenotypic properties of reassortant viruses represent an efficient method for generating a breadth of
information that can inform rapid analysis of surveillance data. However, whether laboratory results
translate to the field strains of interest in nature is uncertain, given differences in the genetic sequences of
the laboratory and field strains and the inherent artificiality of studies conducted in model systems in a
laboratory setting. Characterization of field viruses, an alt-GoF approach, provides direct insight into the
phenotypic properties of reassortant viruses of interest. However, this approach is reactive and depends on
the availability of viral isolates or the publication of a high-quality, complete genome sequence for
synthetic reconstruction of the virus. Additionally, this approach provides limited mechanistic insight into
the relative fitness of reassortant and parental viruses, due to the high genetic diversity among circulating
influenza viruses, and is subject to the same concerns about the translatability of laboratory studies in
model systems as GoF approaches.
The benefit of using experimental data about reassortant viruses (both GoF and alt-GoF) to aid the
interpretation of surveillance data is severely constrained by the quality and availability of existing
genetic surveillance data. Reassortment events are most commonly identified through individual
phylogenetic analysis of each viral gene segment to identify its origin and ancestry. 1724 This requires full
genome sequences and large sequence databases for effective determination of phylogenetic ancestry,
which are not always available, particularly for influenza viruses isolated from animal reservoirs. In
particular, the surveillance of swine populations, thought to play an important role in the generation of
reassortant viruses with pandemic potential due to their ability to be infected with both avian and
1719 Morens DM, Fauci AS (2007) The 1918 influenza pandemic: insights for the 21st century. The Journal of infectious
diseases 195: 1018-1028
1720 Belshe RB (2005) The origins of pandemic influenza--lessons from the 1918 virus. The New England journal of medicine
353: 2209-2211
1721 Worobey M et al (2014) Genesis and pathogenesis of the 1918 pandemic H1N1 influenza A virus. Proc Natl Acad Sci U S A
111: 8107-8112
1723 Smith GJ et al (2009) Dating the emergence of pandemic influenza viruses. Proc Natl Acad Sci U S A 106: 11709-11712

human strains of influenza, is lacking.1725 One factor that contributes to the dearth of high-quality
genetic data to support reassortment analyses is that current diagnostics are not equipped to rapidly
screen and detect reassortants, though several methods have been proposed.1726 Given these
limitations, GoF and alt-GoF approaches to study reassortment currently provide minimal benefits to the
interpretation of surveillance data. Full realization of their potential benefits will require significant
expansion of genetic surveillance for reassortant viruses, particularly in swine populations, which will
pose challenges due to producers’ historical unwillingness to share data with the public health
community.1727
The ability of GoF and alt-GoF approaches to inform assessment of the risk posed by reassortant viruses
detected through surveillance is summarized in Table 15.38. Taken together, both GoF and alt-GoF
approaches provide information about the phenotypic properties of reassortant viruses detected through
surveillance, which can inform analysis of their potential risks to human populations. The proactive
nature of GoF studies facilitates more rapid assessment of surveillance data, but results may not translate
to the strains observed in nature. In contrast, alt-GoF approaches provide more relevant information by
directly studying the surveillance strains of interest but generate information after strains have been
detected and require a viral isolate or high-quality genetic data for synthetic reconstruction of the virus.
However, both approaches currently provide minimal benefits to the interpretation of surveillance data
due to the poor quality of genetic surveillance data for the study of reassortment. Full realization of their
benefits will require significant expansion of surveillance networks, particularly in swine populations, as
well as increasing the quantity of surveillance isolates that are subjected to full genome sequencing.
1725 Vincent A et al (2014) Review of influenza A virus in swine worldwide: a call for increased surveillance and research.
Zoonoses and public health 61: 4-17
1726 Poon LL et al (2010) Rapid detection of reassortment of pandemic H1N1/2009 influenza virus. Clinical chemistry 56: 1340-
1344
1727 (2015l) Swine influenza surveillance. Interview with veterinary influenza researcher.

Table 15.38. Summary of the Benefits of GoF Approaches Involving Reassortment
Benefits to surveillance – Inform evaluation of the risk posed by circulating reassortant viruses detected
through surveillance
GoF Experimental • Efficient generation of a breadth • Results may not translate to field strains
Approaches: of information that can inform of interest in nature
Determination of the analysis of surveillance data o Genetic differences between
laboratory and field strains
reassortment potential • Proactive – generation of
between two viruses and information prior to observation o Artificiality of laboratory experiments
the phenotypic properties of reassortants in nature enables • Limitations in existing genetic
of viable reassortant rapid assessment when similar surveillance data severely constrain the
viruses reassortants emerge ability to detect reassortant viruses
• Reactive – generation of information

following emergence of reassortants in
Alt-GoF Experimental nature
Approaches: • Limited by the availability of viral
Phenotypic • Provides direct insight into the isolates or the publication of high-
characterization of wild phenotypic properties of quality, complete genome sequences for
type reassortant viruses reassortant viruses of interest synthetic reconstruction of viruses
detected through detected through surveillance
surveillance • Limitations in existing genetic
surveillance data severely constrain the
ability to detect reassortant viruses
15.8.5 Benefits to decision-making in public health practice and policy
the results of reassortment studies may stimulate risk mitigation activities to limit the potential for risky
pandemic vaccine development and other pandemic preparedness initiatives. This section discusses the
benefits of GoF approaches relative to alternative approaches for studying reassortment for each of these
areas in turn.
15.8.5.1 GoF benefits to risk mitigation activities that aim to prevent the emergence of reassortant
viruses in nature
could give rise to a reassortant strain comprised of viral gene segments from strains adapted to both
species.1728 GoF approaches that proactively study the reassortment potential between two virus strains
adapted for growth in different species provides insight into reassortants that are viable and that display
1728 (2015k) Interviews with researchers at the National Wildlife Health Center (United States Geological Survey, Department of
the Interior).

phenotypic properties of concern. This information can help to prioritize risk communication about
measures to mitigate the chance of co-infections.1729 For example, hunters would be encouraged to wear
personal protective equipment while gutting birds in areas where avian viruses that are capable of
reassorting with human seasonal viruses are circulating in game bird populations.1730 Another example is
providing guidance to staff and visitors at US National Parks about potentially risky interactions between
people and wildlife and about clinical signs of infection in people, as National Parks provide an unusually
high number of opportunities for humans and wildlife to mix.1731 Researchers at the National Wildlife
Health Center (United States Geological Survey, Department of the Interior), who are often called upon to
provide this kind of “prevention” advice, emphasized that messaging must be targeted and meaningful for
buy-in. Data from GoF reassortment studies provides an evidence base for messaging that may increase
awareness and compliance among the target population. The results of GoF reassortment studies may also
inform biosecurity practices at farms, with respect to interactions between farm workers and animals,
interactions between different species of animals (e.g. poultry and swine at mixed-species farms), and
interactions between agricultural animals and wildlife.
Environmental conditions that provide opportunities for co-infections with a human seasonal virus and an
animal flu virus that has already caused human infections are of high concern regardless of results from
laboratory reassortment studies (i.e. the phenotypic properties of viable reassortants).1732 Thus, GoF
studies that investigate the reassortment potential between human seasonal viruses and animal viruses that
have not yet caused human infections are likely to have a larger impact on public health practice. For
example, many influenza researchers expressed strong interest in understanding whether the highly
pathogenic avian influenza H5N2 virus that caused widespread outbreaks in US domesticated poultry
populations in the summer of 2015 is capable of reassorting with human seasonal viruses.1733 As USDA
experts expect the virus to return this fall, this information could impact risk communication and
recommended biosafety practices for implementing control measures (culling animals, decontaminating
farms, etc.).1734 Because alternative experimental approaches for studying reassortment are reactive, i.e.
limited to studying co-infections and reassortment events that have already occurred in nature, they are
unlikely to be useful for informing public health practices related to reassortment prevention.
Notably, the likelihood that co-infections and subsequent reassortment occurs also depends on complex
ecological factors such as the distribution of viruses within and among reservoir species, which are poorly
understood. An improved understanding of these factors is needed to further refine risk communication
and community-level intervention efforts that aim to prevent the emergence of novel influenza viruses in
human populations through reassortment. These factors can be studied using alternative approaches such
as characterizing the prevalence and distribution of influenza viruses circulating within and between
animal reservoir species, determining the frequency of co-infection events in nature and the parameters
determining outcomes of co-infection, and identifying relevant intermediate hosts. Together, this
information can provide insight into the factors that drive reassortment events in nature.
Taken together, GoF studies that proactively study the reassortment potential between human seasonal
viruses and animal viruses that have not yet caused human infections may help to prioritize risk
communication and risk mitigation measures that aim to limit cross-species interactions that would

1730 (2015k) Interviews with researchers at the National Wildlife Health Center (United States Geological Survey, Department of
the Interior).
1731 Ibid.
1734 USDA issues plan for likely fall return of avian flu. CIDRAP. http://www.cidrap.umn.edu/news-perspective/2015/09/usda-
issues-plan-likely-fall-return-avian-flu. Last Update September 21, 2015. Accessed November 7, 2015.

provide opportunities for co-infection. These data also provide an evidence base for risk mitigation
messaging that may increase compliance among the target population. Alternative approaches can provide
insight into the ecological factors that drive reassortment in nature, which is also needed to refine
prioritization of risk communication and mitigation activities.
As environmental conditions that provide opportunities for co-infections with a human seasonal virus and
an animal virus that has caused human infections are already of high concern, reassortment studies
involving these viruses are unlikely to further increase preventive measures that are already in place.
15.8.5.2 GoF benefits to pandemic risk assessments and downstream decision-making for pandemic
preparedness
pandemic vaccines. The genomic variation risk element of the Influenza Risk Assessment Tool (IRAT)
used by the USG for pandemic risk assessments, described in detail in section 15.3.5.2, includes
consideration of reassortment. Specifically, reassortment between different lineages or sub-types of
viruses raises the risk score for this element. GoF approaches that provide insight into the properties of
reassortant viruses, in particular their fitness, transmissibility, and virulence, could be used to refine the
scores associated with this risk element. In this way, GoF approaches may benefit downstream decision-
making in public health policy. Because viruses that undergo risk assessments are also subjected to
phenotypic characterization of virulence and transmissibility, the main benefit afforded by GoF data is
that it can be generated proactively to enable evaluation of pandemic risk as soon as the genetic sequence
of a virus is published.
In addition to genomic variation, several other types of information related to the properties of the virus
are considered in the risk assessment: phenotypic data (i.e. transmissibility and virulence in ferrets),
epidemiological data (i.e. the number and severity of human infections), and ecological data (i.e. factors
related to infections in animals). In general, the genomic variation risk element is of low- to intermediate-
importance relative to these other factors, though corroboration of phenotypic data adds value to the
assessment by increasing certainty in downstream decisions. However, as discussed in detail in section
15.3.5.2, this risk element may play a relatively more important role in the assessment when a novel virus
first emerges in human populations, if sequences are published prior to the shipping of viral isolates to the
US. The ability to evaluate risk based on genetic sequence data enables a rapid risk assessment, which
may trigger the decision to develop a CVV, providing a head start on vaccine production that would be
valuable in the event of a pandemic.
15.9 Evaluation of the Globalization Potential of GoF Research
15.9.1 Summary of Findings
risk-benefit comparison. For GoF research involving PPPs, the risks are global. This section assesses the
potential for select benefits of GoF research conducted in the U.S. to diffuse globally, in order to inform
the comparison of risks and benefits associated with this research. The potential for three types of GoF
benefits to globalize are considered:
• Improvements to the production of egg- and cell-based influenza vaccines,
• Assistance in the development of new influenza and coronavirus small molecule antivirals, and

• Contributions to risk assessments of circulating animal influenza viruses (pre-pandemic), which
in turn inform prioritization of pandemic preparedness activities such as the development of pre-
pandemic vaccines.
15.9.1.1 Improvements to the production of egg- and cell-based influenza vaccines
Several developing countries have the capacity to directly harness GoF research with potential to benefit
the production of egg- and cell-based influenza vaccines. Specifically, non-high income countries host 18
vaccine producers spanning eight countries, representing an increase in the number of producers and
vaccine-producing countries since 2010. However, the fact that eight out of 13 influenza vaccine
producers that received funding from BARDA contracts allotted in 2006 are not yet marketing an
influenza vaccine highlights the slow timescale for establishing new influenza vaccine production lines in
developing countries. Barriers include human factors (e.g. alleged corruption leading to delays in the
construction of manufacturing facilities), technical factors (e.g. contamination of vaccine), and economic
factors (e.g. lack of domestic demand). Lack of demand for influenza vaccines in-country appears to be a
particularly important issue facing all producers, which is compounded by a lack of knowledge about
optimal vaccination strategies, with respect to vaccine composition and the timing of vaccine delivery, in
tropical regions.
U.S. vaccine donations in the event of a pandemic provide a second pathway for GoF-derived benefits to
reach developing countries. The United States donated approximately 14% of the vaccines committed to
the WHO during the 2009 H1N1 pandemic response, which collectively were deployed to 77 countries.
However, in 2009 both vaccine donation and distribution were significantly delayed, and logistical
challenges associated with vaccine distribution further reduced and/or delayed the quantity of vaccine
doses that reached developing countries’ populations. Although some of these shortcomings have been
addressed in theory by the WHO Pandemic Influenza Preparedness Framework, the ability of the U.S. and
pandemic in the world’s developing countries remains untested.
15.9.1.2 Assistance in the development of novel influenza or coronavirus antivirals
The ability of foreign countries to establish production lines for new antivirals depends not only on their
technical and industrial capabilities but also on their ability to negotiate complex patent issues. In cases
where patent protections do not apply, the actual time needed to initiate commercial production of a U.S.-
designed or commercialized antiviral appears to be in the 1-5 year range. Patent protections do not apply
when a patent is not recognized nationally or is abrogated during a medical emergency, or where the
compound can be sublicensed from the patent owner. However, several companies in developing
countries rapidly activated production of influenza antivirals in less than six months in 2005-2006, when
their governments were preparing for a potential H5N1 pandemic, suggesting that a general lack of
demand for influenza antivirals appears to be keeping globalization in check.
The U.S. demonstrated its willingness to donate influenza antivirals during the 2009 H1N1 pandemic.
However, problems of timeliness of supply compounded issues of suboptimal use in-country. The WHO
Pandemic Influenza Preparedness Framework (developed in 2011) seeks to address timeliness issues but
remains untested.
15.9.1.3 Contributions to pandemic risk assessments of circulating influenza viruses

ability to assess the transmissibility and virulence of viruses in ferrets, data which critically inform
pandemic risk assessments, risk assessments are carried out in collaboration with WHO and laboratory
members of the GISRS (including the CDC). Similar to USG risk assessments, these risk assessments
incorporate information derived from GoF research, alongside epidemiologic and virologic data, and
environmental factors that influence the pandemic potential of the virus.
15.9.2 Introduction
risk-benefit comparison. For GoF research involving PPPs, the risks– that biosafety or biosecurity
In contrast, whether GoF benefits are globally distributed is likely to vary by the type of benefit
considered. The extent to which these benefits can be globalized influences whether risks and benefits are
equally distributed for a particular type of GoF study.
To inform deliberations on this issue, this section evaluates the globalization potential of select GoF
benefits to public health in developing countries. That is, the potential for the outputs of GoF research
conducted in the U.S. to benefit the health of human populations in low- and middle-income bracket
countries, as defined by the World Bank, is assessed.1735
Three types of GoF benefits are considered in this section:
• Benefits to the development and production of egg- and cell-based influenza vaccines,
1735 This classification system is used by the World Health Organization.

July 7, 2015.

• Benefits to the development of new antivirals for influenza viruses or coronaviruses, and
• Benefits to risk assessments of circulating animal influenza viruses (pre-pandemic), which may in
turn stimulate pandemic preparedness activities such as enhanced surveillance and the
development of pre-pandemic vaccines.
Currently, there are no FDA-approved vaccines for MERS-CoV or SARS-CoV.1736,1737 The development
of CoV vaccines is an active area of research, including research into multiple vaccine platforms (e.g.
recombinant vaccines, live attenuated vaccines, DNA vaccines, etc.). GoF research that alters host
tropism and enhances virulence in appropriate animal models has the potential to benefit the development
of CoV vaccines, through the generation of mouse-adapted viruses that serve as a robust animal model for
testing the safety and efficacy of vaccine candidates. However, which type of vaccine will be most rapidly
developed and will prove to be most effective is not clear based on current CoV vaccinology research.
Because the resources and expertise that are required to develop production capacity for different types of
vaccines varies, the globalization potential and barriers to globalization for hypothetical CoV vaccines
cannot be evaluated. Similarly, uncertainty in factors related to the globalization of benefits related to the
development of novel influenza vaccines (derived from GoF approaches that lead to evasion of existing
natural or induced adaptive immunity or that enhance virulence) precludes a meaningful evaluation of the
globalization of these benefits. Thus, the scope of this assessment of the globalization potential of benefits
related to vaccines is limited to GoF benefits to the development and production of existing influenza
vaccines.
The globalization potential for benefits to the development of therapeutics is evaluated based on case
studies on the globalization of production and use of the four influenza antivirals that are currently
licensed in developed countries, which are all small molecule compounds. Therapeutics targeting MERS-
CoV and SARS-CoV are currently in the development phase and include small molecule compounds as
well as other types of therapeutics (e.g. monoclonal antibodies).1738 Setting up hypothetical future
production lines for CoV small molecule antiviral drugs is likely to require a similar level of resources
and expertise as needed for the development of production lines for influenza small molecule drugs. As
such, and in contrast to the evaluation of benefits to vaccine production, this evaluation of the
globalization potential of benefits to therapeutic development applies to relevant research involving CoVs
as well as research involving influenza viruses.
Currently, GoF approaches involving coronaviruses do not have the potential to benefit surveillance.
Although CoV researchers stated that they could envision using information about the molecular
determinants of human adaptation and virulence to assess the risk posed by animal CoVs circulating in
nature, similar to the influenza field, this application is currently unfeasible for two reasons: (1) CoV
surveillance networks are extremely limited, with large gaps in coverage in humans and animals, and (2)
The state of knowledge about the molecular determinants of human adaptation and virulence is poor.1739
Therefore, the analysis of the globalization potential of benefits related to surveillance and pandemic risk
assessments is restricted to research involving influenza viruses.
1736 Centers for Disease Control and Prevention (CDC), “Middle East Respiratory Syndrome (MERS),” June 2, 2015,
http://www.cdc.gov/coronavirus/mers/about/prevention.html. Accessed July 7, 2015.
1737 World Health Organization, “Severe Acute Respiratory Syndrome (SARS),” December 1, 2013,
http://www.who.int/immunization/topics/sars/en/. Accessed July 7, 2015.
1738 During the 2003 SARS-CoV epidemic, Ribavirin was used; however, it “did not appear to have a significant effect,” and a
study of patients treated with Ribavirin indicated “that ribavirin provided no benefit in the resolution of symptoms or
survival.” In: Els Keyaerts, Leen Vijgen, Marc Van Ranst, “Current Status of Antiviral Severe Acute Respiratory Syndrome
Coronavirus Research,” Coronaviruses: Molecular and Cellular Biology, ed. Volker Thiel (Norfolk: Caister Academic
Press, 2007), p. 328.

This assessment evaluates each benefit (i.e. benefits to influenza vaccine production, benefits to the
development of influenza, and benefits to risk assessments of zoonotic influenza viruses) in turn. GoF
benefits to vaccines and therapeutics may globalize in two ways:
• Research results can be applied by third countries, with or without U.S. assistance, to the
development and production of vaccines and antivirals abroad.
• Research results can be applied to the development and production of vaccines and antivirals in
the U.S., to be relinquished for distribution to third countries through the World Health
Organization (WHO) in the event of a pandemic or through non-pandemic assistance programs.
To evaluate the globalization potential of GoF benefits to vaccines and therapeutics, indigenous
capabilities for vaccine and therapeutic production are first described, in order to assess the ability of
developing countries to harness the outputs of GoF research directly. Second, relevant U.S. and WHO
international assistance doctrines and frameworks are described and examples of prior U.S. assistance are
reviewed. Taken together, these two parts enable a qualitative assessment of the degree to which GoF
benefits to PPP vaccines and therapeutics may diffuse globally, as well as the timescale over which those
benefits are expected to internationalize. To evaluate the globalization potential of GoF benefits to
pandemic risk assessments of animal influenza viruses, this section reviews whether and GoF research
contributes to risk assessments conducted in developing countries in which high-risk animal influenza
viruses are circulating as well as the ability of countries to mount responsive pandemic preparedness
activities.
15.9.3 Potential Benefit 1- Improvements in the design and production of vaccines
Several types of GoF research have potential to improve the development and production of egg- and cell-
based influenza vaccines, namely GoF research that enhances virus production that leads to evasion of
therapeutics, that enhances pathogenicity, and that leads to evasion of existing natural or induced adaptive
immunity. Here the benefits of GoF research to influenza vaccine production are briefly summarized. For
a detailed analysis of each GoF benefit, refer to the individual entries in the section devoted to the
benefits of each GoF phenotype above.
GoF research that enhances virus production leads to the generation of higher-yield vaccine backbone
strains and candidate vaccine viruses (CVVs) as well as the identification of genetic markers that enhance
the growth of vaccine viruses. These outputs can benefit vaccine production in two ways: (1) through the
direct use of higher yield vaccine viruses by CVV developers, and (2) through the incorporation of high-
yield markers into existing vaccine viruses, by CVV developers or manufacturers. Use of higher-yield
vaccine viruses shortens vaccine production timelines by increasing the rate of bulk antigen production,
which improves the availability and efficacy of influenza vaccines. Specifically, streamlined vaccine
production processes will translate to faster availability of vaccines during a pandemic and will enable
selection of seasonal strains closer to the start of flu season, reducing the likelihood of vaccine mismatch.
Increasing the yield of vaccine antigen per egg or cell also reduces the manufacturing cost of the vaccine,
which may translate to a lower cost per vaccine dose.
GoF research that enhances pathogenicity may lead to the identification of molecular markers of
enhanced pathogenicity, and GoF research that leads to evasion of therapeutics may lead to the
identification of molecular markers of antiviral resistance. Once validated across many strain contexts,
these markers may be removed from the HA and NA genes of vaccine strains through targeted
mutagenesis, thereby increasing the safety of the vaccine production process.

GoF research that leads to the evasion of existing natural or induced immunity may lead to the
identification of molecular markers for antigenic change and provides insight into the evolutionary
mechanisms driving antigenic drift in nature. This information has potential to improve the strain
selection process in several ways, all of which increase the likelihood that vaccine strains will match
circulating strains during their target flu season. Ultimately, better vaccine match translates to improved
vaccine efficacy, which will mitigate the public health impacts of seasonal influenza epidemics.
For all types of GoF research, these benefits may be harnessed by developing countries through direct
application of GoF research outputs to indigenous influenza production lines, or may benefit developing
countries indirectly through U.S. seasonal and pandemic vaccine donations.
production
As summarized above, GoF research has potential to benefit the development and production of influenza
vaccines through modifications to vaccine strains used for the production of egg- and cell-based vaccines,
which could enhance the safety of the vaccine production process and could improve the quality and
availability of vaccines. High yield CVVs for seasonal and pandemic influenza strains, which serve as the
basis for vaccine strains used for large-scale manufacturing of vaccines, are developed by WHO
Collaborating Centres (WHOCCs) for Influenza and other collaborating laboratories.1740,1741,1742 The
WHO Pandemic Influenza Preparedness Framework stipulates that influenza CVVs be made available
from WHOCCs to any influenza vaccine manufacturer and any other laboratory who makes a request, as
long as the requestor meets appropriate biosafety requirements to receive the strain in question.1743 The
GISRS provides the international framework for the sharing of such biological materials between
laboratories around the world.1744 Vaccine manufacturers then serially passage CVVs to adapt the viruses
for growth in their production systems and further enhance yields, in order to develop vaccine seed strains
that are used for large-scale production of vaccines. Thus, any benefits to strain selection of seasonal
influenza viruses (i.e. GoF research that leads to evasion of existing natural or induced adaptive
immunity), which determine the composition of CVVs, are inherently global. GoF research that leads to
the identification of molecular markers of high-yield, virulence, or antiviral resistance (i.e. GoF research
that enhances virus production, enhances virulence, or leads to evasion of therapeutics) can be applied by
CVV developers or vaccine manufacturers. That is, molecular markers of high growth can be
incorporated in, or molecular markers of antiviral resistance or virulence can be mutated out, of CVVs or
vaccine seed strains by CVV developers or manufacturers, respectively.
Therefore, the ability of developing countries to directly benefit from GoF research conducted in the U.S.
depends on their industrial capacity to produce influenza vaccines. If a developing country has an existing
commercial vaccine production line, the country could either harness GoF benefits through utilization of
modified CVVs, provided by WHOCCs, or through the application of GoF research findings to vaccine
seed strains developed by indigenous manufacturers. Altogether, the likelihood and timescale over which
GoF benefits to vaccine production can be realized depends on two factors: (1) for those countries that do
1740 World Health Organization (WHO), “Influenza: Influenza vaccine viruses and reagents,”
http://www.who.int/influenza/vaccines/virus/en/. Accessed July 7, 2015.
1741 World Health Organization (WHO), “Influenza: Virus Sharing,” http://www.who.int/influenza/pip/virus_sharing/en/.
access to vaccines and other benefits (Geneva: World Health Organization Press, 2011), p. 16-17,
http://apps.who.int/iris/bitstream/10665/44796/1/9789241503082_eng.pdf. Accessed July 7, 2015.
1743 Ibid, p. 12, 15-17.
1744 World Health Organization (WHO), “Global Health Observatory (GHO) data: Global influenza virological surveillance,”
http://www.who.int/gho/epidemic_diseases/influenza/virological_surveillance/en/. Accessed July 7, 2015.

not yet have influenza vaccine production capabilities, the resources needed for and challenges associated
with the establishment of new influenza vaccine production lines, and (2) for those countries that already
have influenza vaccine production capabilities, the country’s regulatory policies governing changes in
vaccine strains. Although an assessment of country-specific regulatory policies as they pertain to the use
of modified vaccine strains is outside the scope of the current study, the FDA does not require regulatory
approval for the commercial use of modified vaccine strains (i.e. there is no regulatory barrier for GoF
benefits to vaccine production in the U.S.).
15.9.3.1.1 Production of influenza vaccines abroad
Global influenza production capacity was most recently comprehensively surveyed in 2010 by the WHO.
The WHO study identified 28 manufacturers that either produced influenza vaccine or were slated to
produce influenza vaccine by 2015.1745 Each manufacturer or potential future manufacturer was then
classified by the World Bank income groups of their home country (simplified to high-, medium-, or low-
income).1746 Of these, 14 manufacturers were in high-income and 14 were in middle-income countries.1747
Based on the reported findings, there were at least 11 vaccines on the market from manufacturers in
middle-income countries in 2010, with at least another 8 vaccines in development.1748 For comparative
purposes, manufacturers based in high-income countries had at least 16 vaccines on the market at the
time, and at least 35 additional vaccines being developed (most using novel technologies).1749
Table 15.39. Summary of Influenza Vaccine Production Capabilities by Country Type, 2010, WHO
data1750
Production method Number of producers in high- Number of Producers in low-income

income countries* countries*
Current In development Current In development
At least
Egg-based production** None At least ten At least eight
thirteen
Cell-based production** Three At least five One None
Other production
None At least 30 None None
methods***
*Some manufacturers produce more than one type of vaccine. As a result, the sum of the number of
manufacturers listed in the table is greater than the total number of manufacturers reported above.
**Includes producers of inactivated and live attenuated vaccines.
***Namely: recombinant haemagglutinin and viral-like particle vaccines (mammalian, insect cells, plant-based,
or other), “universal” vaccines, viral vector vaccines, and DNA vaccines.
The survey identified the following five middle-income countries as having domestic influenza vaccines:
1745 [WHO] Technical Studies Under Resolution WHA63.1, Final Document, A/PIP/OEWG/3/2, April 4, 2011, p. 22-26,
http://apps.who.int/gb/pip/pdf_files/OEWG3/A_PIP_OEWG_3_2-en.pdf. Accessed July 28, 2015.
1746 Ibid.
1747 Ibid.
1748 Ibid.
1749 Ibid.
1750 Ibid.

China, India, Thailand, Indonesia, and Romania.1751,1752 Planned production lines were identified in the
following nine middle-income countries: Brazil, Egypt, Kazakhstan, Mexico, Serbia, South Africa,
Thailand, Iran, and Vietnam.1753
No updated list of active human influenza vaccine manufacturers in 2014 or 2015 has been made publicly
available. A dataset of influenza producers was therefore compiled to compare the current influenza
production situation with that surveyed in 2010. First, we determined which of the 28 companies
identified by the WHO 2010 survey are still currently commercially producing influenza vaccines.1754
This list was supplemented with current or planned influenza manufacturers outside of high-income
countries listed in the Developing Countries Vaccine Manufacturers Network (DCVMN) directories from
2014 and 2015,1755,1756,1757 in the International Federation of Pharmaceutical Manufacturers &
Associations’ Influenza Vaccine Supply Members list,1758 and in the U.S. Department of Health and
Human Services’ Influenza Vaccine International Capacity Building Portfolio.1759 These were
subsequently bolstered by searches on potential manufacturers flagged in the literature or in news
reports.1760 A referenced list is provided in Section 15.9.6, and the findings are summarized in the figure
below.
1751 Marie-Paule Kieny, “Overview of Global and Regional Influenza Vaccine Production Capacity,” presentation given at the
WHO GAP-II Vaccine Production Capacity conference, Geneva, Switzerland, July 13, 2011, p.6,
http://www.who.int/influenza_vaccines_plan/resources/mpk_b.pdf. Accessed October 29, 2015.
1752 Jeffrey Partridge, Marie Paule Kieny, “Global production capacity of seasonal influenza vaccine in 2011,” Vaccine 31, no. 5
(January 2013): p. 728-731, http://www.sciencedirect.com/science/article/pii/S0264410X12015861. Accessed October 1,
2015.
1753 Ibid.
1754 The list of company names provided in note 1 was used. Jeffrey Partridge, Marie Paule Kieny, “Global production capacity
of seasonal influenza vaccine in 2011,” Vaccine 31, no. 5 (January 2013): p. 728-731,
http://www.sciencedirect.com/science/article/pii/S0264410X12015861. Accessed October 1, 2015.
1755 DCVMN is a coordinating platform for vaccine producers in the developing world. Note that certain DCVMN producers are
in countries that are currently classed by the World Bank as being High-Income countries (such as South Korea).
For a description of the DCVMN, see: Sonia Pagliusi et al., “Developing Countries Vaccine Manufacturers Network: Doing
good by making high-quality vaccines affordable for all,” Vaccine 31 supplement 2 (April 2013): p. B176-B183,
http://www.sciencedirect.com/science/article/pii/S0264410X1201701X. Accessed July 13, 2015.
1756 Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p.1-96,
http://www.dcvmn.org/IMG/pdf/directory.pdf. Accessed November 15, 2015.
http://www.dcvmn.org/IMG/pdf/dcvmn_directory_2014.pdf. Accessed July 7, 2015.
1758 International Federation of Pharmaceutical Manufacturers & Associations (IFPMA), “IFPMA Influenza task force – IVS
Membership,” http://www.ifpma.org/resources/influenza-vaccines/ifpma-influenza-task-force/ivs-membership.html.
1759 U.S. Department of Health & Human Services, “International Influenza Vaccine Capacity Building Portfolio,”
https://www.medicalcountermeasures.gov/projectmaps/Who.aspx. Accessed October 1, 2015.
1760 Jan Hendriks, Yan Liang, Bing Zeng, “China’s emerging vaccine industry,” Human Vaccines 6, no. 7 (2010): p. 602-607,
http://www.tandfonline.com/doi/pdf/10.4161/hv.6.7.11933. Accessed October 29, 2015

Figure 15.3. Developing countries that host at least one company with an influenza vaccine currently on the
market are shaded in deep blue. Developing countries that host at least one company with R&D efforts for
the production of an influenza vaccine are shaded in light blue.
Analysis of the assembled dataset reveals that the number of active producers outside of high-income
countries has increased since 2010. In total, 18 companies in middle-income countries were found to be
actively producing influenza vaccines, compared to 11 manufacturers in 2010.1761 These were mostly
located in China (8) and India (4), although Bangladesh, Brazil, Indonesia, Iran, Kazakhstan, and Pakistan
each had one producer. One country has stopped production, since the sole Romanian manufacturer
marked as active in 2010 has stopped marketing influenza vaccines.1762 At least 13 additional companies
have R&D work for influenza vaccines at various stages of completion. These companies were located in
China (4), Egypt (1), Iran (2), Serbia (1), Thailand (3), and Vietnam (2).
As many of the new influenza vaccine manufacturers between 2010 and 2015 are located in countries that
already had influenza vaccine production capabilities, overall the geographic distribution of production
capacities outside of high-income countries has only moderately expanded. Eight countries now produce
influenza vaccines (up from five). Based on current R&D efforts, an additional five countries may
become influenza vaccine producers in the future.1763
1761 This count excludes companies based in Taiwan, as the World Bank classes “Taiwan, China” as a “high-income” economy,
separately from “China,” which it classes as an “upper-middle-income” economy. See:
July 7, 2015.
1762 “Institutul Cantacuzino nu face vaccin antigripal nici in sezonul 2015 - 2016, desi are autorizatii” [Cantacuzino Institute will
not make flu vaccine in the 2015-2016 season, despite having licenses], Ziare, May 21, 2015,
http://www.ziare.com/social/spital/institutul-cantacuzino-nu-face-vaccin-antigripal-nici-in-sezonul-2015-2016-desi-are-
autorizatii-1364363. Accessed October 1, 2015.
1763 Namely: Egypt, Iran, Serbia, Thailand, and Vietnam.

A lack of end-user demand appears to be a recurring and common problem that is preventing several of
the middle-income firms mentioned in this section from initiating or maintaining influenza vaccine
production. With respect to pandemic influenza vaccines, this issue stems from a lack of government
support to purchase vaccines for pandemic preparedness purposes. For example, representatives of the
Serum Institute of India argued that the Indian government’s decision not to purchase an H1N1 vaccine it
had initially financially supported had “threatened the sustainability of influenza production capacity in
India” and had resulted in six million doses of unsold vaccine.1764 With respect to seasonal influenza
vaccines, this issue involves a lack of demand by individuals. For example, a presentation by a senior
advisor on disease control from the Ministry of Public Health of Thailand on vaccine production plans in-
country noted that there simply had been no demand for seasonal influenza vaccine before the 2004 H5N1
outbreak affected the country.1765 Notably, the Chinese market experience has demonstrated that domestic
demand for seasonal influenza vaccine increases with the income level of individuals, thus low domestic
demand is to be expected outside of high income countries.1766 This demand issue is compounded by the
fact that current recommendations for the strain composition of seasonal influenza vaccines are geared
toward countries in the Northern and Southern hemispheres with well-defined flu seasons, such as the
United States and Australia.1767 In contrast, well-defined seasonality does not always occur in tropical
regions of the world; instead, low levels of influenza virus circulate throughout the year. In these regions,
optimal vaccination strategies, including whether Northern or Southern hemisphere vaccines are more
protective and when during the year vaccines are best deployed, are not well understood. Research to
better understand patterns of influenza transmission and seasonality in the tropics, as well as how best to
mitigate the public health burden associated with influenza through vaccination, is ongoing. This research
provides an important foundation for developing countries’ efforts to bolster their vaccine production
capabilities and increase in-country demand in the future.
15.9.3.1.2 U.S. vaccine production assistance
Several U.S. programs seek to support the aforementioned ability of developing countries to produce
vaccines. Since seasonal vaccine production lines are adapted to produce pandemic vaccines, these
pandemic preparedness programs complement seasonal influenza production assistance, and vice
versa.1768
The U.S. HHS supports production capabilities abroad for seasonal and pandemic influenza vaccine
through funding provided by its Biomedical Advance Research and Development Authority (BARDA)
branch.1769 For example, BARDA provided funding to enable Vietnam-based VABIOTECH’s planned
1764 World Health Organization (WHO), “Report of the Sixth Meeting with International Partners on Prospects for Influenza
Vaccine Technology Transfer to Developing Country Vaccine Manufacturers,” p.15,
http://apps.who.int/iris/bitstream/10665/85515/1/9789241505994_eng.pdf. Accessed August 3, 2015.
1765 Suwit Wibulpolprasert, “GAP and Flu Vaccine Production in Thailand – from Public Health Policy Development to Vaccine
Production,” presentation given at the Second WHO Consultation on the Global Action Plan for Influenza Vaccine (GAP-
II), Geneva, Switzerland, July 12, 2010, p.6, http://www.who.int/influenza_vaccines_plan/resources/suwit.pdf. Accessed
October 1, 2010.
1766 Eliza Yibing Zhou, “Vaccine Development in China,” BioPharm International 20, no. 4 (April 2007): p.1,
http://www.biopharminternational.com/china-today-vaccine-development-china. Accessed October 29, 2015.
1768 For U.S. context, see: Executive Office of the President, President’s Council of Advisors on Science and Technology,
[U.S.A.] “Report to the President on Reengineering the Influenza Vaccine Production Enterprise to Meet the Challenges of
Pandemic Influenza,” August 2010, https://www.whitehouse.gov/sites/default/files/microsites/ostp/PCAST-Influenza-
Vaccinology-Report.pdf. Accessed July 7, 2015.
1769 PATH, “PATH’s Work in Vaccine Development: Low-cost influenza vaccine production,”
http://sites.path.org/vaccinedevelopment/influenza/vaccine-production-in-the-developing-world/. Accessed August 3, 2015.

cell-based influenza vaccine production capacity.1770,1771 Overall, BARDA has provided financial support
to 13 companies in 12 medium-income countries seeking to develop influenza vaccine production
lines.1772,1773 According to the U.S. HHS Public Health Emergency database on BARDA’s portfolio, the
contracts that provided funding to these firms were all awarded in September 2006.1774
As of mid-2014, BARDA had provided “more than $50 million in cooperation with WHO” distributed in
the form of grants to potential vaccine producers in developing countries to assist them in setting up
influenza vaccine production lines.1775 BARDA further provided “over $20 million to support vaccine
adjuvant technology transfer, biomanufacturing workforce training, and clinical trial and manufacturing
technical support to developing country influenza vaccine manufacturers.”1776
Of the 13 companies that received support from BARDA, six appear to remain in the R&D phase, one has
ceased production of vaccines, one appears to have halted R&D efforts, and five currently produce
influenza vaccines. The status, future plans, and reasons for production delays are summarized in the
following table.
Table 15.40. Summary of the Status of Influenza Vaccine Production Companies in Developing
Countries that Received Funding from BARDA in 2006.
Company Country Status of Vaccine Future Plans Reasons for Production
Production Delays
Production facility
was under
construction, but
Birmex has stopped
Acera de Birmex1777 Mexico listing an influenza Unknown Unknown
vaccine under its
DCVMN 2015
product R&D
description1778
Vaccine Technology Transfer to Developing Country Vaccine Manufacturers,” Dubai, United Arab Emirates, March 18-19,
2013, p. 8, http://who.int/iris/bitstream/10665/85515/1/9789241505994_eng.pdf. Accessed August 3, 2015.
1771 Centers for Disease Control and Prevention (CDC), “Influenza Division International Activities, Fiscal Years 2012 & 2013
Annual Report,” p. 121, http://www.cdc.gov/flu/pdf/international/program/2012-2013-intl-program-report.pdf. Accessed
August 3, 2015.
1772 These companies are: Acera de Birmex (Mexico), BCHT (China), BioFarma (Indonesia), Cantacuzino Institute (Romania),
GPO (Thailand), Instituto Butantan (Brazil), IVAC (Vietnam), RIBSP (Kazakhstan), Serum Institute of India (India), The
BioVac Institute (South Africa), Torlak Institute (Serbia), VABIOTECH (Vietnam), and VACSERA (Egypt).
1773 U.S. Department of Health & Human Services, “International Influenza Vaccine Capacity Building Portfolio.”
1774 Ibid.
1775 United States of America, “Report on USA implementation of Article X of the Biological and Toxin Weapons Convention,”
Meeting of the States Parties to the Convention on the Prohibition of the Development, Production and Stockpiling of
Bacteriological (Biological) and Toxin Weapons and on Their Destruction, Meeting of Experts, Geneva, Switzerland,
August 4-8, 2014, BWC/MSP/2014/MX/INF.5, p.4 para. 10. Accessed July 7, 2015.
1776 Ibid.
1777 Luis Guillermo F. Ibarra PL et al., “Influenza Vaccine Project at Birmex,” poster presented at the Eighth Meeting with
International Partners on Prospects for Influenza Vaccine Technology Transfer to Developing Country Vaccine
Manufacturers, Sao Paulo, Brazil, March 17-18, 2015, http://www.who.int/phi/8thPartnersMtg2015_Birmex_poster.pdf.
1778 Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p. 47-48,

Production Delays
Production facility
construction is Begin clinical
BCHT1779 China N/A
complete; vaccine trials in 2015
R&D is ongoing
R&D for fill-finish
operations (final stage
Expect to
of production) was
obtain a license
ongoing in March
for
South 2015; as of November
BioVac Institute1780 domestically N/A
Africa 2015 the company
filling seasonal
lists an influenza
vaccine in
vaccine under its
20161782
DCVMN 2015
marketed products1781
1779 Jinchang Wu, “Changchun BCHT Biotechnology Co., China,” poster presented at the Eighth Meeting with International
Partners on Prospects for Influenza Vaccine Technology Transfer to Developing Country Vaccine Manufacturers, Sao
Paulo, Brazil, March 17-18, 2015, http://www.who.int/phi/8thPartnersMtg2015_BCHT_poster.pdf. Accessed November 5,
2015.
1780 Patrick Tippoo, Simphiwe Ntombela, “The BIOVAC Institute: Establishing Influenza Vaccine Manufacturing Capacity in
Africa,” poster presented at the Eighth Meeting with International Partners on Prospects for Influenza Vaccine Technology
Transfer to Developing Country Vaccine Manufacturers, Sao Paulo, Brazil, March 17-18, 2015,
http://www.who.int/phi/8thPartnersMtg2015_BIOVAC_poster.pdf. Accessed November 5, 2015.
1781 Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p. 75-76,
1782 Patrick Tippoo, Simphiwe Ntombela, “The BIOVAC Institute: Establishing Influenza Vaccine Manufacturing Capacity in
Africa,” poster presented at the Eighth Meeting with International Partners on Prospects for Influenza Vaccine Technology
http://www.who.int/phi/8thPartnersMtg2015_BIOVAC_poster.pdf. Accessed November 5, 2015.

Production Delays
Industrial plant
Obtained EUA for
construction started in 2009;
H1N1 vaccine in
corruption investigations led
GPO1783,1784,1785 Thailand 2011; construction of Unknown
to suspension of funds ;
industrial production
construction re-approved in
plant ongoing
20141786,1787,1788,1789,1790
Ceased active Withdrew vaccines from the
production of market due to low antigen
Cantacuzino
Romania influenza vaccines for Unknown concentration in 2012 and
Institute1791
the 2015 – 2016 due to endotoxin
season1792 contamination in 20131793
Conducted a Phase I
clinical trial for their
IVAC1794 Vietnam Unknown N/A
A/H5N1 vaccine in
2014 – 2015.1795
Preclinical trials for
Torlak1796,1797 Serbia seasonal flu vaccine Unknown N/A
are ongoing
1783 Somchaiya Surichan et al., “Development of influenza vaccine production capacity by the Government Pharmaceutical
Organization of Thailand: Addressing the threat of an influenza pandemic,” Vaccine 29 Supplement 1 (July 2011), p. A29-
A33.
1784
Government Pharmaceutical Organization, “Our Products,” http://www.intergpomed.com/Default.aspx?tabid=61. Accessed
November 5, 2015.
1786 “Vaccine factory to restart construction,” Bangkok Post, December 11, 2014,
http://www.bangkokpost.com/lite/news/448902/vaccine-factory-to-restart-construction. Accessed November 5, 2015.
1787 Eric Palmer, “Thailand government vaccine plant at center of probe,” Fierce Pharma Manufacturing, April 25, 2013,
http://www.fiercepharmamanufacturing.com/story/thailand-government-vaccine-plant-center-probe/2013-04-25. Accessed
November 5, 2015.
1788 Pongphon Sarnsamak, “Flu Vaccine Plant Saraburi: DSI Agrees to Look Into Irregularities,” The Nation, April 12, 2013,
retrieved at: http://www.thaivisa.com/forum/topic/632500-flu-vaccine-plant-saraburi-d-s-i-agrees-to-look-into-
irregularities/. Accessed November 5, 2015.
1789 Puangchompoo Prasert, Piyanut Thamnukasetchai, “Paracetamol Scandal: Action Sought against top GPO official,” The
Nation, May 2, 2013, retrieved at http://www.thaivisa.com/forum/topic/636661-paracetamol-scandal-action-sought-against-
top-thai-official/. Accessed November 5, 2015.
1790 Suriyan Panyawai, “GPO chief’s axing ‘not political’: Board’s probe was thorough, chairman says,” Thailand Online News,
May 20, 2013, http://onlinenewsthailand.com/2013/05/20/gpo-chiefs-axing-not-political/. Accessed November 5, 2015.
1791 “Institutul Cantacuzino nu face vaccin antigripal nici in sezonul 2015 - 2016, desi are autorizatii,” [Cantacuzino Institute
will not make flu vaccine in the 2015-2016 season, despite having licenses], Ziare, May 21, 2015,
http://www.ziare.com/social/spital/institutul-cantacuzino-nu-face-vaccin-antigripal-nici-in-sezonul-2015-2016-desi-are-
1792 Ibid.
1793 Ibid.
1794 “Influenza A/H5N1 Vaccine Clinical Trial (IVACFLU-A/H5N1) - Phase 1,” ClinicalTrials.gov, October 15, 2015,
https://clinicaltrials.gov/ct2/show/record/NCT02171819. Accessed October 29, 2015.
1795 Ibid.
1796 Torlak, “History,” http://www.torlakinstitut.com/en/page/23/History. Accessed October 29, 2015.
1797 Torlak, “Research & Development,” http://www.torlakinstitut.com/en/page/14/Research+&+Development. Accessed
October 29, 2015.

Production Delays
Lack of market demand for
Completed Phase III
pandemic vaccine (reliance
clinical trials for their Filed for a
VABIOTECH1798,1799 Vietnam on government purchases
cell-based A/H5N1 license in 2013
discourages private
vaccine in 2012
investors)
Fill-finished first
batches in 2008;
obtained licensing for
the product in 2009
Bio Farma1800 Indonesia N/A N/A
(first licensed product
resulting from WHO
tech transfer
program)1801
First domestic batch
produced in 2011;
obtained certificate of
Instituto Butantan1802 Brazil good production N/A N/A
practices from
national regulator in
20121803
Pre-pandemic H5N1
Research Institute for and H1N1 vaccines
Biological Safety Kazakhstan registered with the N/A N/A
Problems (RIBSP)1804 government in
2013.1805, 1806
1798 VABIOTECH, “Products – Vaccine,”

http://www.en.vabiotech.com.vn/index.php?option=com_content&view=article&id=88&Itemid=109&lang=en. Accessed
October 29, 2015.
1799 Juliet Bryant, “Influenza vaccine manufacturing in Viet Nam: Report on the APACI Satellite session,” One Health, 2015,
http://onehealth.org.vn/influenza-vaccine-manufacturing-in-viet-namreport-on-the-apaci-satellite-session.new. Accessed
October 29, 2015.
1801 Mahendra Suhardono, Dori Ugiyadi, Ida Nurnaeni, Imelda Emelia, “Establishment of pandemic influenza vaccine
production capacity at Bio Farma, Indonesia,” Vaccine 39, supplement 1 (July 2011): p. A22-A25,
<http://www.sciencedirect.com/science/article/pii/S0264410X1100689X>.
1802 Butantan Institute, “Butantan Institute Influenza Vaccine Production,” poster presented at the Eighth Meeting with
International Partners on Prospects for Influenza Vaccine Technology Transfer to Developing Country Vaccine
Manufacturers, Sao Paulo, Brazil, March 17-18, 2015, <http://www.who.int/phi/8thPartnersMtg2015_Butantan_poster.pdf>.
1803 Marcelo De Franco, Jorge Kalil, “The Butantan Institute: History and Future Perspectives,” PLoS Neglected Tropical
Diseases 8, no. 7 (July 2014): p. e2862, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4080994/pdf/pntd.0002862.pdf>.
1804 Research Institute for Biological Safety Problems (RIBSP), “Technology transfer project for Influenza Vaccine-2011/14
phase,” poster presented at the Eighth Meeting with International Partners on Prospects for Influenza Vaccine Technology
<http://www.who.int/phi/8thPartnersMtg2015_RIBSP_poster.pdf>.
1805 Ibid.
1806 “Реестр отечественных поставщиков товаров фармацевтической и медицинской промышленности” [Register of
domestic suppliers of goods Pharmaceutical and Medical Industries], March 13, 2015,
<http://arkalyk.kostanay.gov.kz/uploads/files/1d03853ecd302518be6a42de19ca184a.doc>.

Production Delays
Pandemic H1N1
vaccine licensed July
2010; subsequently
Serum Institute of
India created a trivalent N/A N/A
India Ltd. 1807
seasonal influenza
vaccine currently on
the market.1808, 1809
In conclusion, more than eight years after BARDA began its assistance program, roughly two thirds of the
funding recipients appear to lack an influenza vaccine product on the market. Since the method by which
the funding recipients were picked has not been made public, it is unclear whether the company case
studies truly represent the average capability of vaccine companies in middle-income developing
countries to set up new production lines. What can be concluded from the case studies is that the four
success cases demonstrate that some developing countries are able to develop, produce, and market a new
influenza vaccine given eight years. However, the human, technical, and economic problems encountered
by the other companies drive home the point that setting up new influenza vaccine production lines is
time-consuming and is a high-risk endeavor from a business perspective.
15.9.3.2 U.S. vaccine donations
The United States supports foreign seasonal and pandemic influenza vaccine stockpiles through direct
vaccine donations, which represents a different pathway for the globalization of GoF benefits related to
vaccine development and production. Specifically, any GoF-derived improvements to U.S. vaccine
development and production will indirectly benefit developed countries that receive U.S.-produced
vaccines through assistance and emergency response programs.
15.9.3.2.1 U.S. seasonal vaccine donations
The U.S. HHS’ Centers for Disease Control has recently begun donating seasonal vaccines in an effort to
increase seasonal influenza vaccination in developing countries.
The U.S. CDC organizes the donation of seasonal influenza vaccines as part of the vaccine donation
portion of the Partnership for Influenza Vaccine Introduction.1810 A first donation cycle was conducted in
2012, whereby 375,000 doses of vaccine donated by Walgreens Company (U.S.A.) led to the vaccination
of 355,902 individuals in the Lao People’s Democratic Republic.1811 The program was expanded in 2013,
with programs launched in Nicaragua and Uganda. Lao received 100,000 doses, and Nicaragua 35,000, in
1808 Ibid.
1809 F. Marc LaForce, “Developing a Trivalent Live Attenuated Influenza Vaccine,” presentation given at the Workshop on
Business Modeling for Sustainable Influenza Vaccine Manufacturing, Washington, D.C., U.S.A., January 14-16, 2013,
<http://www.who.int/influenza_vaccines_plan/resources/session_5_laforce.pdf>.
1810 The Task Force for Global Health, “Partnership for Influenza Vaccine Introduction,” <http://www.taskforce.org/our-
1811 Joseph Bresee, CDC, “Global Action Plan for Influenza Vaccines – II: CDC’s Supportive Activities,” GAP-II Partners
Meeting, Dubai, United Arab Emirates, March 18, 2013,
<http://www.who.int/phi/Day1_9_Bresee_GAP2_CDC_PM_Dubai2013.pdf>.

2013.1812 Additional private donors that donated vaccines, supplies, or subsidized shipping services
included bioCSL, Becton Dickinson and Company, and UPS.1813 At the U.S. national level, DOD
provided assistance, in particular through the donation of 5,000 vaccine doses from US Air Force bases in
Kadena, Japan.1814
Several factors significantly limit the impact of this program. First, the program relies on private
donations from manufacturers which in turn are “based on [the] availability of excess vaccine supply” and
are therefore unpredictable and potentially limited.1815 Second, WHO guidelines stipulate that the vaccine
must be licensed for use in the recipient country.1816 The amount of time necessary for the initial license
of a seasonal influenza vaccine in-country will vary by country but is generally a lengthy process (e.g. 10
months in the U.S.).1817 As this timeframe is too long for a given seasonal vaccine donation to be licensed
in time for flu season, vaccine donations must be matched with countries that already have the vaccines
approved for use. However, since the countries that would benefit most from vaccine donations do not
have domestic influenza production capabilities and weak public health systems, including regulatory
infrastructure for MCMs, many lack approval for available influenza vaccines.1818 And finally, there is a
problem of timing, as the correct hemisphere vaccine (Northern or Southern) must be donated at the right
time to match the recipient country’s influenza season, which limits donation options.1819 This timing
issue is compounded by late commitment announcements.1820 Since donators currently donate vaccine
surplus, they can most easily provide stocks after the U.S. influenza season, but this may be too late for
potential recipient countries with similar influenza seasons. As a result, the range of countries that can
receive U.S. donations under these programs is greatly limited.
15.9.3.2.2 U.S. vaccine donations in response to a pandemic
In the event of a pandemic, U.S. national policy calls for donations of vaccines to the WHO for
redistribution to developing countries. As a member state to the WHO Pandemic Influenza Preparedness
Framework, the U.S. is committed to supplying influenza vaccines to a WHO-maintained pandemic
benefit-sharing system, which would then redistribute vaccines to developing countries as necessary to
respond to a pandemic.1821 The U.S. HHS is the lead agency for the relinquishing of assets to international
organizations in response to an outbreak. It, “in coordination with other United States Government
Agencies, responds to requests for assistance from foreign countries and international organizations by
contributing available HHS expertise and assets, including personnel and medical countermeasures (e.g.
vaccines, antivirals and diagnostics).”1822
The exact quantity to be contributed by each member state is left for members to decide in the event of a
1812 Alan R. Hinman, “Partnership for Influenza Vaccine Introduction (PIVI),” Dubai, United Arab Emirates, March 25, 2014,
p.2, <http://www.who.int/phi/DAY1_08_Panel2_Hinman_Panel2_PIVI_PM_Dubai2014.pdf>.
1813 Centers for Disease Control and Prevention (CDC), “Laos and Nicaragua Protect High-Risk Persons from Influenza, with
Help from Donor Coalition and CDC,” <http://www.cdc.gov/flu/international/highlight-high-risk.htm>.
1814 The Task Force for Global Health, “Partnership for Influenza Vaccine Introduction (PIVI),” <http://www.taskforce.org/our-
1816 Ibid.
Vaccine Technology Transfer to Developing Country Vaccine Manufacturers,” p.21,
<http://apps.who.int/iris/bitstream/10665/85515/1/9789241505994_eng.pdf>.
1819 Ibid.
1820 Ibid.
1822 U.S. Department of Health & Human Services, “North American Plan For Animal and Pandemic Influenza (NAPAPI),”
April 2, 2012, p. 16.

pandemic; the WHO’s guidance document calls for the provision of an “appropriate contribution to this
system.”1823 The WHO guidance document however makes clear that the vaccine donations should be
structured as a percentage of vaccine production runs, to ensure timely supply.1824 The following case
study, on the U.S. vaccine donations in response to the 2009 H1N1 pandemic, show how and to what
extent U.S. vaccine donations can reach developing countries.
The 2009 pandemic preceded and motivated the formation of the WHO’s Pandemic Influenza
Preparedness Framework in 2011. As such, although the actions taken by the U.S. during the pandemic
remain instructive, certain shortcomings in the international donation and response system have been
addressed by the establishment of a Framework.
15.9.3.2.3 Case study: U.S. pandemic vaccine donations during the 2009 H1N1 pandemic
During the H1N1 influenza pandemic, U.S. vaccine donations were organized in response to 17 bilateral
requests and a call for “global solidarity” from the WHO Director General.1825 In September 2009, the
United States pledged up to 10% of its vaccine production runs to the WHO; eight other countries
subsequently made similar pledges.1826,1827 The U.S. H1N1 influenza response established a “10%” rule of
thumb, whereby 10% of vaccine production runs would be donated to the WHO for distribution to
developing countries in need of assistance.
The decision to relinquish vaccines to the WHO for international deployment was coordinated by the
White House Security Staff International H1N1 Vaccine Assistance Working Group across several U.S.
agencies (i.e., this required more than HHS input).1828 HHS has described the decision process as
depending upon, inter alia:
• Vaccine supplies, in particular supplies available for international deployment,

• Domestic need and demand,
• Requests from WHO and bilateral requests,
• Legal authority to procure and deploy the vaccines,
• Available funding,
• The quantity and source of the required ancillary supplies, and
• Options for financing transportation and deployment.1829
WHO served as the overall coordinator during the donation process, but USAID assisted in the
development of country vaccination plans and with carrying out the necessary vaccination campaigns.1830
In total, the United States donated 16,860,100 doses of 2009 H1N1 influenza vaccine to the WHO for
international distribution, which represented approximately 14% of the vaccines committed to the
1824 Ibid.
<http://www.phe.gov/Preparedness/mcm/h1n1-retrospective/Documents/h1n1-retrospective.pdf>.
1826 The eight countries were: Australia, Brazil, France, Italy, New Zealand, Norway, Switzerland, and the United Kingdom.
1827 World Health Organization, “Report of the WHO Pandemic Influenza A(H1N1) Vaccine Deployment Initiative,” 2012, p. 4,
<http://www.who.int/influenza_vaccines_plan/resources/h1n1_deployment_report.pdf>.
1828 Ibid.
1829 Ibid.
1830 Ibid.

WHO.1831,1832 Out of a total of 122,450,000 vaccine doses committed by all states, the WHO distributed a
total of 78,066,290 doses of vaccines to 77 countries.1833
Overall, donation of vaccines to WHO suffered from severe timeliness issues. Vaccine production and
domestic supply difficulties in the U.S. (and other developed countries) in turn impacted vaccine
donations. In October 2009, limited vaccine availability forced the U.S. HHS Secretary to publicly
announce that the U.S. would delay the promised vaccine donations until the slated at-risk population in
the U.S. could be vaccinated.1834,1835 (Notably, production of the vaccine was delayed due to difficulties in
generating a high-yield vaccine strain that was suitable for large-scale production, a shortcoming that GoF
research that enhances virus production aims to address.) Similar delays in promised deliveries could
occur again, if future pandemic strains generate similarly low-yield vaccine strains. Advanced purchase
agreements, whereby a given number of vaccines not yet produced are purchased by a government from a
private vaccine producer, compounded accessibility issues.1836 Since the vaccines already belonged to a
particular buyer, the private firm was unable to donate a portion of the run to the WHO, regardless of a
desire to do so.1837
Other developed countries were reticent in donating vaccines, and in a particularly severe pandemic
whether promised doses would reach developing countries in time to be effective is unclear.1838 For
example, Canada’s 5 million vaccine dose donation began only after the second wave of the flu pandemic
was declared “over” in-country.1839 Several developed countries– such as France, Germany, Switzerland,
and the Netherlands– tried to sell excess vaccines instead of donating them.1840,1841 For example, when
only roughly 5 million people in France accepted the vaccine out of a stockpile order of 94 million doses,
France attempted to sell its stocks at the same price it had obtained the vaccines.1842 The WHO Pandemic
Influenza Preparedness Framework’s explicit clause on the provision of vaccines on a rolling basis seeks
1833 The commitment of vaccines to the WHO involves a signed agreement, and therefore goes beyond a political pledge.
World Health Organization, “Final Pandemic (H1N1) 2009 Vaccine Deployment Update,” November 10, 2010,
<http://www.who.int/csr/disease/swineflu/action/h1n1_vaccine_deployment_final_update_2010_11_10.pdf>.
1834 David P. Fidler, Kelly Lee, “Negotiating Equitable Access to Influenza Vaccines: Global Health Diplomacy and the
Controversies Surrounding Avian Influenza H5N1 and Pandemic Influenza H1N1,” PLoS Med 7, no. 5 (May 2010),
<http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2864298/>.
One 7, no. 3 (2012), <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3295778/>.
1836 Sam F. Halabi “Obstacles to pH1N1 Vaccine Availability: The Complex Contracting Relationship among Vaccine
Manufacturers, the World Health Organization, Donor and Beneficiary Governments,” The Public Health Response to 2009
H1N1: A Systems Perspective, eds. Michael A. Stoto, Melissa A. Hidgon (New York: Oxford University Press, 2015), p.
207.
1837 Ibid.
1838 David P. Fidler, Kelley Lee, “Negotiating Equitable Access to Influenza Vaccines: Global Health Diplomacy and the
Controversies Surrounding Avian Influenza H5N1 and Pandemic Influenza H1N1.”
One 7, no. 3 (2012), <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3295778/>.
1840 Ibid.
1841 “La France veut revendre ses vaccins contre la grippe A,” [France wants to sell its vaccines against influenza A] Le
Parisien, January 3, 2010, <http://www.leparisien.fr/societe/la-france-veut-revendre-ses-vaccins-contre-la-grippe-a-03-01-
2010-763246.php>.
1842 Ibid.

to prevent this particular donation timeliness problem, but whether countries will comply with the
Framework during a severe pandemic remains untested.1843
In addition to delays in the donation of vaccine doses, the planning and execution of the donation and
distribution of vaccine doses and ancillary supplies was hampered by several factors that further delayed
and/or reduced the quantity of vaccine doses distributed to recipient countries. The process by which U.S.
vaccines were donated and reached the end-users was negatively affected by “liability issues, vaccine
registration requirements, and ensuring that recipient countries had in place funding and approved
Vaccine Deployment and Vaccination Plans to support distribution of the vaccine.”1844 WHO had to
coordinate the vaccination plan in the recipient country with the U.S. donor, and HHS diplomatically
noted the existence of some U.S.-WHO coordination friction by stating: “HHS and USAID […] remained
in close contact in order to coordinate the deployment of vaccine, transport of vaccine and the deployment
of ancillary items, though ultimate decisions on the recipient countries were made by the WHO based on
their allocation procedures—thus, these decisions were not necessarily aligned.”1845 Since the nature and
extent of these disagreements have not been revealed, it is difficult to ascertain their impact on the
timeliness of donated vaccine availability.
In conclusion, roughly 14% of the WHO distributed vaccines during the H1N1 pandemic were donated by
the United States, which collectively reached 77 recipient countries. The pandemic response suffered
from serious timeliness issues with donations, coupled with logistical challenges during distribution and
during in-country vaccination. These challenges highlight that, while US donation of vaccines is a viable
pathway by which GoF benefits to vaccine production may globalize, the time needed to orchestrate the
logistics of vaccine shipment and vaccination in-country will delay delivery of a vaccine to a developing
country’s population relative to a scenario in which that country is capable of indigenously producing and
freely distributing its own vaccine doses.
GoF research has potential to benefit the production of vaccines in several ways: (1) through the
development of higher-yield vaccine viruses, which shorten vaccine production timelines to enhance the
availability and efficacy of vaccines, (2) through improving strain selection capabilities for seasonal
influenza vaccines, which improves vaccine efficacy by increasing the likelihood of vaccine match, and
(3) through the identification of molecular markers for enhanced virulence and antiviral resistance, which
can be removed from vaccine strains to enhance the safety of the vaccine production process. These
benefits can be realized by developing countries in two ways: (1) through the direct application of GoF
research insights to production in-country, and (2) through the receipt of U.S.-produced vaccines donated
through assistance or emergency response programs.
With respect to indigenous production capabilities, both the total number of vaccine producers outside of
high-income countries (17) and the number of non-high income producing countries (7) has increased
since 2010. As WHOCCs provide ready access to candidate vaccine strains to all such producers, these
countries are currently capable of harnessing GoF research benefits to vaccine production. The total
number of producers outside of high-income countries is slated to increase by as many as an additional
six countries given current R&D efforts by at least 13 companies spanning a total of eight non high-
income countries. Analysis of the R&D timelines for foreign influenza vaccine manufacturers that
1844 World Health Organization, “Final Pandemic (H1N1) 2009 Vaccine Deployment Update,” November 10, 2010,
<http://www.who.int/csr/disease/swineflu/action/h1n1_vaccine_deployment_final_update_2010_11_10.pdf>.
1845 Ibid.

received BARDA funding support in 2006 shows that bringing a new influenza vaccine to market may
require up to eight years.
This analysis revealed significant challenges associated with the establishment of new influenza vaccine
production lines in developing countries, as only four of the 13 producers that received funding in 2006
are currently actively producing influenza vaccines. Impediments to the establishment of production lines
include human factors (e.g. alleged corruption delaying construction of manufacturing facilities),
technical factors (e.g. contamination of vaccine doses), and economic factors (e.g. lack of domestic
demand). Lack of demand for influenza vaccines in-country appears to be a particularly important issue
facing all producers, which is compounded by a lack of knowledge about optimal vaccination strategies,
with respect to vaccine composition and the timing of vaccine delivery, in tropical regions. Therefore,
whether current R&D efforts for the establishment of new production lines will come to fruition is
uncertain, and the rate of continued development of new production capabilities in the future cannot be
ascertained.
U.S. donations of pandemic or seasonal flu vaccines provide a second pathway for GoF-derived benefits
to reach developing countries. The U.S. experience during the 2009 H1N1 pandemic demonstrated that,
although the U.S. was committed to providing some 10% of its vaccine stocks to developing countries
through the WHO, the effectiveness of these donations suffered from serious timeliness issues. Although
the WHO Pandemic Influenza Preparedness Framework (developed in 2011) established guidelines for
vaccine donation during a pandemic in an effort to address these shortcomings, the ability of the U.S. and
pandemic in the world’s developing countries remains unverified. The U.S. CDC organizes the donation
of surplus seasonal influenza vaccines from vaccine manufacturers to developing countries, but several
factors significantly limit the impact of this program, including the need for donated vaccines to be
licensed in the recipient country and mismatches between the timing of vaccine availability and the needs
of recipient countries.
15.9.4 Potential Benefit 2- Assistance in the development of new influenza or coronavirus antivirals
Several types of GoF research have the potential to inform the development of new influenza or
coronavirus antivirals, namely GoF research that alters host tropism, that enhances pathogenicity, and that
leads to evasion of antivirals. Here we briefly summarize how GoF research outputs benefit the
development of new therapeutics. For a detailed analysis of each GoF benefit, refer to individual benefit
sections devoted to the benefits of each GoF phenotype above.
First, GoF approaches that enhance the virulence of influenza viruses or coronaviruses may lead to the
identification of novel virulence factors that are good therapeutic targets, thereby enabling the
development of novel therapeutics.
Second, GoF approaches that alter the host range of CoVs enable the development of mouse-adapted
virus strains. Unlike other animal models for CoVs, infection of mice with mouse-adapted CoV strains
mimics the pathology of human disease; thus mouse-adapted strains serve as a robust system for testing
the safety and efficacy of candidate therapeutics. Notably, because mouse-adapted strains are the only
model system that satisfies the FDA Animal Efficacy Rule, the use of mouse-adapted strains is essential
for the licensure of new CoV therapeutics in the US.
Third, GoF approaches that lead to evasion of therapeutics inform the development of new therapeutics
for influenza viruses and coronaviruses. Specifically, these approaches provide insight into the
mechanism of action of therapeutics and demonstrate the genetic threshold for acquisition of resistance

(i.e. the number of mutations that are required to gain resistance), which speaks to the potential field
efficacy of the therapeutic. Both types of data are recommended for inclusion in an Investigational New
Drug (IND) application to the FDA, thus this approach also plays a critical role in the licensure of new
therapeutics in the US.
These benefits may be harnessed by developing countries either through indigenous production of new
antivirals, or through direct U.S. donations of antivirals in the event of a pandemic.
15.9.4.1 Capacity for foreign production of GoF-derived new influenza antivirals
Several developing countries produce antivirals against influenza that were originally developed in high-
income countries, including the U.S. Several countries also conduct research on novel influenza antiviral
candidates originally discovered in developed countries.
The process by which a pharmaceutical company abroad can proceed to produce an antiviral compound
discovered in the U.S. is complex. When a novel compound showing medical promise is developed into a
potential treatment by scientists working for a company, the company typically owns the rights to the
discovery as per the scientists’ contracts, and is then free to patent the potential treatment. For instance,
scientists at Gilead Sciences discovered what would become the influenza antiviral medication Tamiflu,
and Gilead Sciences held the patent on Tamiflu.1846
Since patents are filed at the national level, certain U.S. pharmaceutical companies’ compounds are not
patent protected in certain countries that nevertheless have domestic antiviral pharmaceutical production
capabilities. For instance, Tamiflu is not patent protected in Thailand, the Philippines, and Indonesia.1847
Pharmaceutical companies based in these countries are therefore free to produce the underlying active
compound of Tamiflu (oseltamivir) as a generic medication, provided that no additional bilateral or
multilateral trade agreement clauses prohibits this activity.
For countries where a U.S. patent is legally valid or where a U.S. invention has been patented in-country,
domestic producers can either obtain a license or challenge the patent’s validity by producing the
compound without a license. The licensing process allows a patent holder to include limits and conditions
that it otherwise could not impose by simply selling the patent, such as the requirement to sell the product
within specific geographic confines (such as a single country).1848 Gilead Sciences, for example, licensed
their compound’s patent to the pharmaceutical company Roche as part of a co-development agreement for
Tamiflu signed in 1996; the amended license agreement text reportedly allows Gilead Sciences to play a
role in Roche’s oversight of the compound’s manufacture and commercialization and its pandemic
planning for the product.1849,1850 In practice, however, firms are often reluctant to license production in
order to maintain production line exclusivity, and governmental and public pressure has played a role in
convincing U.S. firms to grant licenses to foreign companies. Roche was for instance threatened by
several Congress representatives with a temporary abrogation of the Tamiflu patent when the firm was
unable to meet demand during the 2005 H5N1 pandemic preparedness period, after which the company
reached a number of sub-licensing agreements with other companies abroad to produce the compound.1851
1846 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,” CRS Report for Congress, August 16, 2007, p. 7,
retrieved at <http://www.ipmall.info/hosted_resources/crs/RL33159_070816.pdf>.
1848 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,” p. 7.
1849 Ibid.
1850 Gilead Sciences Inc., “Gilead and Roche End Tamiflu® Dispute; Expanded Collaboration Includes Gilead Role in Oversight
of Manufacturing and Commercialization,” November 16, 2005,
<http://investors.gilead.com/phoenix.zhtml?c=69964&p=irol-newsArticle&ID=783456>.
1851 Brian T. Yeh, “Influenza Antiviral Drugs and Patent Law Issues,” p. 3-4.

Indeed, national patent law traditionally allows governments to cancel medication patents or to force the
licensing of the compounds in response to medical emergencies.1852 In cases where the applicability of a
patent is disputed or unclear and where there is a potential emergency need, governments may simply
decide not to attempt to enforce patent laws. The chairman of the Indian company Cipla alluded to such a
situation when he declared in 2005, at the height of the shortage issues around the antiviral, that: “Right
or wrong, we're going to commercialize and make oseltamivir.”1853
Patents protect a product for a significant period of time. The first U.S. patent covering Tamiflu, for
instance, was filed in 1996 by Gilead Sciences, and the company is still fighting in court attempts to
produce generic oseltamivir medication by referencing its patent protections.1854,1855 Once associated
patents on a compound and its manufacturing expire, all competitors are allowed to produce the
compound as a generic medication.1856
The following section focuses on the ability of foreign countries to establish production lines for notional
novel influenza or coronavirus antivirals developed in the United States with assistance from GoF
research. Deriving benefits from such a U.S. discovery, however, goes beyond the foreign country’s
ability to establish a production line. It will also crucially depend on its ability to negotiate the complex
patent issues noted above.
Current patent and licensing laws are in a state of flux, as a result of growing public and governmental
pressure for affordable medication at the national level, and as a result of comprehensive multinational
trading negotiations that would potentially make it easier for pharmaceutical companies to obtain patents
and increase the ability of companies to sue governments over intellectual property losses.1857 The
following section draws lessons from recent cases of globalization of antiviral production lines, but these
conclusions reflect the current policy landscape and may become less relevant if patenting and licensing
laws significantly change in the future.
15.9.4.1.1 Capacity for novel influenza antiviral production abroad
This section considers the capacity of developing countries to establish production lines for new antivirals
developed with assistance from GoF research. The experience with the globalization of production
capabilities for the existing influenza antivirals zanamivir, oseltamivir, and peramivir (approved for use in
the U.S.), as well as for laninamivir octanoate (approved for use in Japan) are used as case studies to
estimate the length of time needed to establish production of a new antiviral. Of note, all four antivirals
are small molecule compounds, and all were discovered in high-income (developed) countries.
As of 2015, zanamivir, oseltamivir, and peramivir, but not laninamivir octanoate, were approved for use
1852 Donald G. McNeil Jr., “Indian Company to Make Generic Version of Flu Drug Tamiflu,” The New York Times, October 14,
2005, <http://www.nytimes.com/2005/10/14/health/indian-company-to-make-generic-version-of-flu-drug-tamiflu.html>.
1853 Ibid.
1854 Kali Hays, “Gilead Sues Lupin Over Plans To Produce Generic Tamiflu,” Law 360, September 17, 2015,
<http://www.law360.com/articles/703920/gilead-sues-lupin-over-plans-to-produce-generic-tamiflu>.
1855 U.S. Patent 5,763,483 A, “Carbocyclic Compounds,” Filed December 27, 1996, Published June 9, 1998,
<http://www.google.com/patents/US5763483>.
1856 World Health Organization (WHO), “Generic Drugs,” <http://www.who.int/trade/glossary/story034/en/>.
1857 “Hard pills to swallow,” The Economist, January 4, 2014, <http://www.economist.com/news/international/21592655-drug-
firms-have-new-medicines-and-patients-are-desperate-them-arguments-over>.

against influenza in the United States.1858,1859 With the exception of the newly-discovered peramivir and
laninamivir octanoate compounds, these antivirals were all listed in the 2004 WHO Guidelines on the Use
of Vaccines and Antivirals During Influenza Pandemics.1860 Table 15.41 below summarizes information
on these antivirals.
1858 Centers for Disease Control and Prevention (CDC), “Influenza Antiviral Medications; Summary for Clinicians,” p.1,
retrieved at “Antiviral Drugs: Recommendations of the Advisory Committee on Immunization Practices (ACIP):
Information for Health Care Professionals,” March 4, 2015, <http://www.cdc.gov/flu/pdf/professionals/antivirals/antiviral-
summary-clinician.pdf>.
1859 Ribavirin is mentioned in the literature but its effectiveness has been questioned. See:
World Health Organization, “WHO Guidelines on the Use of Vaccines and Antivirals during Influenza Pandemics,”
WHO/CDS/CSR/RMD/2004.8, 2004, Annex 5, p. 3,
<http://www.who.int/csr/resources/publications/influenza/11_29_01_A.pdf>.
1860 World Health Organization, “WHO Guidelines on the Use of Vaccines and Antivirals during Influenza Pandemics,” Annex
5, p. 3.

Table 15.41. Information on influenza antivirals
Generic Proprietary Brand Category Year Earliest FDA approval, any formulation
name manufacturer 1861 name compound
published
Neuraminidase
Zanamivir GlaxoSmithKline Relenza 1993.1862 July 1999.1863
inhibitors
Neuraminidase
Oseltamivir Roche Tamiflu 1997.1864 October 1999.1865
inhibitors
Neuraminidase
Peramivir Biocryst Rapivab 2000.1866 Emergency use in 2009, approved for use in December 2014. 1867
inhibitors
Laninamivir Biota Pharmaceuticals Neuraminidase Currently not FDA-approved; approved for use in Japan against
Inavir 2009.1868
octanoate and Daiichi Sankyo inhibitors Influenza A and B since 2010 and 2013, respectively. 1869
1861 [WHO] Technical Studies Under Resolution WHA63.1, Final Document, A/PIP/OEWG/3/2, p. 117;
“Biota Reports That Laninamivir Octanoate is Approved for the Prevention of Influenza in Japan,” Biota, December 20, 2013,
<http://investors.biotapharma.com/releasedetail.cfm?releaseid=815483>.
1862 Mark Von Itzstein et al., “Rational Design of potent sialidase-based inhibitors of influenza virus replication,” Nature 363 (June 1993): p. 418-423,
<http://www.nature.com/nature/journal/v363/n6428/abs/363418a0.html>.
1863
U.S. Food and Drug Administration, “FDA Approved Drug Products: Drug Details, RELENZA,”
<http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Search.DrugDetails>.
1864 Kim C. U. et al., “Influenza neuraminidase inhibitors possessing a novel hydrophobic interaction in the enzyme active site: design, synthesis, and structural analysis of
carbocyclic sialic acid analogues with potent anti-influenza activity,” J. Am. Chem Soc. (January 1997): p. 681-690, <http://www.ncbi.nlm.nih.gov/pubmed/16526129>;
1865 U.S. Food and Drug Administration, “FDA Approved Drug Products: Drug Details, TAMIFLU,”
<http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm?fuseaction=Search.DrugDetails>.
1866 Babu Y.S. et al., “BCX-1812 (RWJ-270201): discovery of a novel, highly potent, orally active, and selective influenza neuraminidase inhibitor through structure-based drug
design,” Journal of Medical Chemistry 43, no. 19 (2000): p. 3482-3486.
1867 U.S. Food and Drug Administration, “FDA approves Rapivab to treat flu infection,” FDA News Release, December 22, 2014,
<http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm427755.htm>.
1868 Makoto Yamashita et al., “CS-8958, a Prodrug of the New Neuraminidase Inhibitor R-125489, Shows Long-Acting Anti-Influenza Virus Activity,” Antimicrobial Agents and
Chemotherapy 53, no. 1 (2009): p. 186-192.
1869 Biota Pharmaceuticals, Inc., “Biota Provides Update on BARDA Contract for Laninamivir Octanoate,” May 8, 2014,
<http://investors.biotapharma.com/releasedetail.cfm?releaseid=846423>.

All four compounds have been produced by some middle-income developing countries. Since companies
mostly do not report on R&D efforts nor publicize the terms regarding technology transfers of
sublicenses, finding out the average length of time necessary to establish production capability for a given
degree of technology assistance is very difficult. Efforts to develop production capabilities in developing
countries can nevertheless be broadly grouped into three strategies: licensed activities coupled with
follow-on research, independent ventures, and exploratory research. Some examples of companies in
middle-income countries are given below for each strategy to qualitatively illustrate the challenges and
timescale associated with each approach, although limited details are available for some cases.
Licensed activities coupled with follow-on research
Vietnam received permission from Roche to encapsulate the oseltamivir compound on November 9,
2005.1870,1871 Vietnamese scientists, such as members of the Ha Noi University of Pharmacy and the
Institute of Chemistry of the Vietnamese Institute of Science and Technology, have since been engaged in
laboratory production of oseltamivir, and have also experimented with recycling the substance from
expired tablets.1872,1873,1874
In China, the Shanghai Pharmaceutical Group and HEC Pharm Co. are the two companies licensed to
supply the Chinese state with oseltamivir.1875,1876 Under a restriction imposed by Roche, the producers can
“only use it for pandemic purposes within China”; in practice, the firms were not allowed to sell the
compound commercially and had to furnish oseltamivir to the state at regulated prices.1877 Shanghai
Pharmaceutical Group announced they could produce 200,000 doses in six months when they obtained
their licensing agreement in December 2005.1878 The amount of R&D time invested by the firm prior to
December 2005 to establish this oseltamivir production capacity was not revealed, but the announcement
came some eight years after oseltamivir was identified as a potential MCM in the published literature
(1997).1879
1870 “Calls for more money as the threat looms ever larger,” The Economist, November 11, 2005,
<http://www.economist.com/node/5134571>.
1871 Vietnam Ministry of Foreign Affairs, “Viet Nam signs agreement on Tamiflu production with F.Hofmann-Laroche,” August
10, 2005, <http://www.vietnamembassy-tanzania.org/en/vnemb.vn/tinkhac/ns051111100413>.
1872 “Vietnam likely to produce Tamiflu from anise next year,” Xinhua through People, March 21, 2006,
<http://en.people.cn/200603/21/eng20060321_252323.html>.
1873 “Viet Nam to Produce Tamiflu from Star Aniseed,” Talk Vietnam, March 24, 2006,
<http://www.talkvietnam.com/2006/03/viet-nam-to-produce-tamiflu-from-star-aniseed/>.
1874 “Scientists hope to recycle 10m out-of-date Tamiflu tablets,” Viêt Nam News, August 10, 2015,
<http://vietnamnews.vn/social-issues/health/203702/scientists-hope-to-recycle-10m-out-of-date-tamiflu-tablets.html>.
<http://uk.reuters.com/article/2009/04/30/us-flu-drugs-generic-idUKTRE53T0UL20090430>.
1876 “Roche licenses China firm to produce Tamiflu,” China Daily, December 12, 2005, p.1-2,
<http://www.chinadaily.com.cn/english/doc/2005-12/12/content_502758.htm>.
1877 Roche opens Tamiflu to outside firms,” Swiss Info, December 12, 2005, <http://www.swissinfo.ch/eng/roche-opens-tamiflu-
to-outside-firms/4900404>.
1878 Wang Xu, “Shanghai firm wins license for generic version of Tamiflu,” China Daily, December 13, 2005,
<http://www.chinadaily.com.cn/english/cndy/2005-12/13/content_502775.htm>.
1879 Kim C. U. et al., “Influenza neuraminidase inhibitors possessing a novel hydrophobic interaction in the enzyme active site:
design, synthesis, and structural analysis of carbocyclic sialic acid analogues with potent anti-influenza activity,” J. Am.
Chem Soc. (January 1997): p. 681-690, <http://www.ncbi.nlm.nih.gov/pubmed/16526129>.

Also in China, the firm Nanjing Simcere Dongyuan Pharmaceutical Co. Ltd., a subsidiary of Simcere
Pharmaceutical Group, obtained a license to produce and sell zanamivir in September 2006.1880,1881
According to a Simcere spokesman, GlaxoSmithKline licensed the production of the drug but only
provided “limited technical support” in its synthesis.1882,1883,1884 A pathway was developed in-country
through joint research with the Shanghai Institute of Materia Medica and the Nanjing EffectPharm Drug
Development Corporation.1885 The firm obtained approval from the Chinese national regulator to
manufacture and sell the compound in China in 2010, and the firm is currently selling the compound.1886
The firm has conducted innovative research for production lines related to zanamivir, since the Shanghai
Institute of Materia Medica published papers in 2012 and 2013 on the design and synthesis of zanamivir
analogs.1887,1888
In India, Hetero Drugs obtained a sublicense for the production and sale of oseltamivir in December
2005.1889 The company has since supplied millions of tablets of oseltamivir to the Indian
government.1890,1891
Independent ventures
As noted above, the Indian company Cipla publicly announced in October 2005 that it would
independently produce oseltamivir without entering into a commercial agreement with Roche.1892 In a
subsequent interview, the company chair declared that the company had begun researching oseltamivir
production techniques in 2004.1893 In a parallel effort, Cipla announced it was producing zanamivir
without entering into a commercial agreement with GlaxoSmithKline in 2006.1894,1895 In India today,
Cipla Ltd., Ranbaxy Laboratories, Strides Arcolab, and Natco Pharma all have production capacity for
1880
GlaxoSmithKline, “Agreement to increase availability of Zanamivir supply in Asia and Lease Developed Countries,” May
15, 2007, <http://www.gsk-china.com/asp/News/client/newconten/515200791555.htm>.
1881 PR Newswire, “Simcere Receives SFDA Approval to Manufacture and Sell Zanamivir in China,” Bloomberg, February 11,
2010, <http://www.bloomberg.com/apps/news?pid=21070001&sid=aRO5.9_34evg>.
1882 Ibid.
1883 “Scientists develop ways producing anti-bird flu drug Zanamivir,” People’s Daily, February 6, 2009,
<http://en.people.cn/90001/90781/90878/6587151.html>.
1884 EffacttPharm, “Research Progress,” July 10, 2015, <http://www.effecttpharm.com/yifang_e.html>.
1885 Shanghai Institute of Materia Medica, Chinese Academy of Sciences, “The New Drug Certificate for Anti-H1N1 Flu
Medicine Zanamivir granted to SIMM,” March 17, 2010, <http://english.simm.cas.cn/rp/201003/t20100317_51500.html>.
1886 Simcere, “Zanamivir,” <http://www.simcere.com/english/products/detail.asp?gongs_id=59&leibieid=APIs>.
1887 Feng E. et al., “Structure-based design and synthesis of C-1 and C-4-modified analogs of zanamivir as neuraminidase
inhibitors,” Journal of Medicinal Chemistry 56, no. 3 (2013): p. 671-684.
1888 Ye. D. et al., “Synthesis of C-4-modified zanamivir analogs as neuraminidase inhibitors and their anti-AIV activities,”
European Journal of Medical Chemistry 54 (2012): p. 7640-770.
1889 “Roche grants Tamiflu licence to Hetero Drugs,” The Times of India, December 24, 2005,
<http://timesofindia.indiatimes.com/business/india-business/Roche-grants-Tamiflu-licence-to-Hetero-
Drugs/articleshow/1344422.cms>.
1890 Khomba Singh, “Hetero bags mega chunk of govt’s anti-flu drug deal,” The Economic Times, May 5, 2009,
<http://articles.economictimes.indiatimes.com/2009-05-05/news/27636779_1_hetero-drugs-anti-flu-drug-oseltamivir>.
1891 Gireesh Chandra Prasad, “Govt to buy bird flu drugs from Roche, Hetero,” The Economic Times, December 7, 2005,
<http://articles.economictimes.indiatimes.com/2005-12-07/news/27487189_1_hetero-drugs-bird-flu-task-force>;
1892 “The Tamiflu Manufacturing Controversy: An Interview with Yusuf Hamied,” Multinational Monitor vol. 27, no. 2,
March/April 2006, <http://www.multinationalmonitor.org/mm2006/032006/interview-hamied.html>.
1893 Ibid.
1894 Ibid.
1895 “Cipla MD favours compulsory licensing sans monopoly,” The Hindu Business Line, November 15, 2005,
<http://www.thehindubusinessline.com/todays-paper/tp-corporate/cipla-md-favours-compulsory-licensing-sans-
monopoly/article2195410.ece>.

oseltamivir without having entered into an agreement with Roche.1896,1897,1898,1899,1900
Thailand took advantage of the fact that Tamiflu had not been patent-protected in-country and has had
independent production capacity for the generic oseltamivir since 2006.1901,1902,1903 The Governmental
Pharmaceutical Organization manufactured 200,000 tablets in early February 2006, following an
announcement that it would do so in December 2005.1904
Independent exploratory research
A number of research groups in developing countries publish research on synthesis pathway optimization
for newly discovered antiviral compounds. The ultimate objective of this type of research may be to
prepare for in-country industrial production of the antiviral in question, although end-use intent cannot be
definitely predicted based on publications in the scientific literature.
The chemical compound peramivir (first published in 2000 and approved for emergency use in the US in
2009 and for general use in 2014) has already been synthesized in a novel process by a Chinese research
team, which achieved this result by March 2012 at the latest.1905 Unlike earlier publications that described
known pathways to obtain peramivir that were funded through grants for basic research projects on new
drugs,1906,1907 the Chinese research team developed a new pathway designed for industrial production.
This new research effort was funded by the Guangdong Production and Research Joint Project, which
may indicate an interest in future Chinese domestic production of the compound.1908 In a peer-reviewed
paper that appeared in 2013, the team reported an improved synthetic route for peramivir synthesis with a
total 34% yield that obviated the need for a highly toxic chemical in the final step.1909 At a minimum, the
publication’s results demonstrate that domestic production of the compound is well within China’s
1896 “Resistant strain of swine flu feared; virus killing thousands in India,” Japan Times, February 26, 2015,
<http://www.japantimes.co.jp/news/2015/02/26/asia-pacific/science-health-asia-pacific/resistant-strain-of-swine-flu-feared-
virus-killing-thousands-in-india/#.VcjIdfnZViY>.
1897 “Swine flu: Hetero Healthcare increases Fluvir production by 400%,” The Economic Times, February 26, 2015,
<http://articles.economictimes.indiatimes.com/2015-02-26/news/59541921_1_swine-fluvir-oseltamivir>.
1898 Khomba Singh, “Govt curbs sale of flu drug Zanamivir,” The Economic Times, August 29, 2009,
<http://articles.economictimes.indiatimes.com/2009-08-29/news/28483297_1_swine-flu-drug-oseltamivir-zanamivir>.
<http://uk.reuters.com/article/2009/04/30/us-flu-drugs-generic-idUKTRE53T0UL20090430>.
1900 “Ranbaxy to supply oseltamivir capsules to US,” The Economic Times, October 21, 2007,
<http://articles.economictimes.indiatimes.com/2007-10-21/news/28461984_1_capsules-domestic-sales-generic-version>.
1901 “Tamiflu- Oseltamivir Production,” News Medical, February 1, 2011, <http://www.news-medical.net/health/Tamiflu-
Oseltamivir-Production.aspx>.
1902 Pennapa Hongthong, “Scientists produce generic Tamiflu,” The Nation, August 4, 2006,
<http://www.nationmultimedia.com/2006/08/04/national/national_30010320.php>.
<http://fpc.state.gov/documents/organization/65759.pdf>.
1905 Fei Jia, Juan Hong, Ping-Hua Sun, Jian-Xin Chen, Wei-Min Chen, “Facile Synthesis of the Neuraminidase Inhibitor
Peramivir,” Synthetic Communications: An International Journal for Rapid Communication of Synthetic Organic Chemistry
43, no. 19 (2013): p. 2641-2647, <http://www.tandfonline.com/doi/abs/10.1080/00397911.2012.729279>.
1906 顾轶娜,林东海,“新型抗流感病毒神经氨酸酶抑制剂帕拉米韦研究进展,”中国生化药物杂志 30, no. 4 (2009): p.273-276
[GU Yi-na, LIN Dong-Hai, “Research progress on peramivir as a novel anti-influenza virus neuraminidase inhibitor,”
Chinese Journal of Biochemical Pharmaceutics 30 no. 4 (2009): p.273-276.].
1907 贾飞, 陈良柱, 陈建新, 孙平华, 陈卫民,“帕拉米韦合成路线图解,”中国医药工业杂志 42 no. 12 (2011): p. 954-956. [JIA
Fei, CHEN Jianxin, SUN Pinghua, CHEN Weimin, “Graphical Synthetic Routes of Peramivir,” Chinese Journal of
Pharmaceuticals 42, no. 12 (2011): p. 954-956.].
1908 Fei Jia et al., “Facile Synthesis of the Neuraminidase Inhibitor Peramivir,” p. 2646.
1909 Ibid, p. 2641.

technical capabilities. The peramivir case is one in which a novel synthetic pathway for a U.S. designed
chemical was rapidly developed abroad, indeed even before the compound was approved for general use
in the U.S. by the FDA.
Similarly, in December 2014, a Chinese research team working out of the State Key Laboratory of
Bioorganic and Natural Products Chemistry of the Shanghai Institute of Organic Chemistry published a
novel synthetic pathway for the production of laninamivir octanoate.1910 This published process used an
inexpensive acid as a starting compound, and reportedly obtained a 72% total yield with a 12-step process
that was suitable for scale-up.1911 Overall, the paper’s process effectively lowers industrial production
costs while minimizing losses in yield and minimizing the number of additional industrial steps required:
three extremely important factors necessary for industrial scale-up.
As demonstrated by these accounts, indigenous zanamivir and oseltamivir production lines exist in
several middle-income countries. Several Chinese research groups have also demonstrated the capability
to efficiently synthesize peramivir and laninamivir octanoate, raising the possibility that in-country
production lines could be rapidly set up should the decision to do so be made. Although the amount of
R&D time invested by each of the companies and research teams named above to achieve their
production capability is unknown (i.e. when the company began researching synthetic pathways and/or
began setting up production facilities), conservative estimates demonstrate that at least some middle-
income countries achieved the capacity for full-scale production of a given MCM less than ten years after
the compound was initially published in the literature. In the case of laninamivir octanoate, a Chinese
laboratory demonstrated a novel chemical synthesis process for the compound less than 5 years after the
compound was published in the literature. Notably, several companies rapidly activated production
capabilities capable of producing hundreds of thousands of doses in less than six months in 2005-2006
when their governments were preparing for a potential H5N1 pandemic. This suggests that, as with
influenza vaccines, a general lack of demand for influenza antivirals appears to be keeping production
line globalization in check. Based on these cases, the actual time needed to initiate commercial production
of an antiviral designed in a developed country appears to be in the 1-5 year range.
In conclusion, should GoF research enable the development of a new promising small molecule antiviral
compound targeting influenza viruses or coronaviruses, the experience with current antiviral compounds
suggests that at least some developing countries have the will and the means to develop methods for
production of a potential MCM in-country. In turn, this production capability could then be scaled-up to
industrial production once the compounds can be legally produced either through license or as generics,
improving global pandemic response capabilities. We note that although barriers to the establishment of
production lines may vary between different types of therapeutics (e.g. small molecule drugs versus
monoclonal antibodies), patenting and licensing issues are likely to be the same for all types of
therapeutics.
15.9.4.2 U.S. antiviral donations
GoF benefits to the development of novel antivirals may also globalize through U.S. donations of
antivirals to developing countries. Current U.S. government assistance to antiviral supply abroad are
primarily limited to plans for donations to the WHO for redistribution to developing countries in case of
an influenza pandemic. As a member state in the WHO Pandemic Influenza Preparedness Framework, the
United States government is committed to contributing influenza antivirals to the WHO-organized
Pandemic Influenza Preparedness Benefit Sharing System, which would redistribute MCMs to third
1910 Tian J. et al., “Organocatalytic and scalable synthesis of the anti-influenza drugs zanamivir, laninamivir, and CS-8958,”
Angewandte Chemie 126 (2014): p. 14105-14108.
1911 Ibid, p. 14105-14106.

countries as part of a pandemic response as needed.1912 U.S. private pharmaceutical companies can and
have donated antiviral treatments to the WHO and to countries dealing with local outbreaks independently
from government contributions.1913,1914 However, these private companies are under no obligation to do so
in the future, and hence the effect of this potential GOF-derived benefits dissemination pathway cannot be
reliably assessed.
As there are no licensed therapeutics for coronaviruses in the U.S. or abroad, neither the U.S. nor the
WHO have formal policies or plans in place for the donation of (notional) therapeutics in the event of an
epidemic caused by a novel coronavirus.
The following case study reviews U.S. donations of antivirals to foreign countries during the 2009 H1N1
pandemic and identifies bottlenecks that may pose a barrier to the globalization of GoF benefits via this
pathway in the future. Although the creation of the WHO Pandemic Influenza Preparedness (PIP)
Framework in 2011 limits the extent to which this case study is predictive of the successes and challenges
of influenza antiviral donation efforts in the future given its plan for a joint pre-pandemic influenza
antivirals stockpile,1915 similar challenges could be encountered in the event of ad hoc donation of CoV
therapeutics during a CoV epidemic.
15.9.4.3 Case study: U.S. antiviral donations during the 2009 H1N1 pandemic
The comprehensive after-action report, “An HHS Retrospective on the 2009 H1N1 Influenza Pandemic to
Advance All Hazards Preparedness,” does not expound on the U.S. decision to disburse antivirals to other
nations beyond noting that it was carried out by HHS “after careful consideration of federal policies and
discussions of global demand.”1916 Responding to the 2009 H1N1 pandemic was initially an ad-hoc
process, given the lack of a uniform U.S. decision-making process at the start of the pandemic. The need
for such a process proved to be a major lessons-learned from the pandemic. Given that a U.S. national
framework for decision-making has since been developed, this shortcoming is less likely to hamper
international donation and distribution of influenza antivirals in the event of a future pandemic. This
national framework applies to all “international requests for public health emergency medical
countermeasures”1917 and hence would also apply during a hypothetical CoV pandemic.
The U.S. initially gave 400,000 antiviral treatment courses to Mexico, followed by 420,000 courses of
oseltamivir for the Pan American Health Organization.1918 The Pan American Health Organization then
provided stocks to countries throughout Latin America and the Caribbean.1919 Although this demonstrates
U.S. willingness to provide antiviral doses in the event of a pandemic, one U.S. public health policy
stakeholder stated that the global health security enterprise may not be as willing to donate antivirals in
1913 David Reddy, “Responding to pandemic (H1N1) 2009 influenza: the role of oseltamivir,” J. Antimicrob. Chemother. 65
supplement 2 (April 2010): ii35-ii40, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2835510/pdf/dkq014.pdf>.
access to vaccines and other benefits (Geneva: World Health Organization Press, 2011), p. 18,
<http://apps.who.int/iris/bitstream/10665/44796/1/9789241503082_eng.pdf>.
1916 “An HHS Retrospective on the 2009 H1N1 Influenza Pandemic to Advance All Hazards Preparedness,” p. 38.
1917 Public Health Emergency, U.S. Department of Health & Human Services, “International Assistance and Response Policy
Branch,” October 16, 2014, <http://www.phe.gov/about/OPP/dihs/Pages/policy.aspx>.

the event of future pandemics due to the expense associated with storing and deploying the drugs.1920
The use of donated antivirals during the H1N1 pandemic in developing countries was in general
suboptimal, in part due to the low availability of the antiviral compounds.1921 In Asia for instance, an
authoritative review article noted that “health practitioners were reluctant to follow the recommendation
of the empiric use of oseltamivir”: the practitioners did not wish to use scarce doses on ostensibly mild
cases of influenza, even when the patient was in a high-risk group.1922
In sum, although U.S. policy supports the donation of influenza antivirals in the event of a pandemic, the
relatively small number of doses donated in comparison to the global need in the event of a pandemic
means that developing countries would face shortages, which would in turn exacerbate poor usage in-
country.
GoF research has the potential to benefit the development of novel therapeutics for influenza viruses and
coronaviruses in several ways. First, GoF research that enhances virulence enables the identification of
novel virulence factors, which may be good therapeutic targets. Second, mouse models for CoVs,
developed through GoF approaches that alter host range, represent a robust system for testing the safety
and efficacy of therapeutics in development. Third, GoF approaches that lead to evasion of therapeutics
also support the licensure of new therapeutics by providing information that is critical for an
The ability of developing countries to establish production lines for novel influenza or hypothetical
coronavirus therapeutics depend not only on their manufacturing capabilities but also on their ability to
negotiate the complex patent issues surrounding the marketing of therapeutics. In cases where patent
protections do not apply, analysis of the timeline and circumstances surrounding the establishment of
production lines for existing influenza antivirals in several developing countries suggest that the time
needed to initiate commercial production of a U.S.-designed or commercialized antiviral is one to five
years. Patent protections do not apply when a patent is not recognized nationally or is abrogated during a
medical emergency, or where the compound can be sublicensed from the patent owner. Notably, several
companies in developing countries rapidly activated influenza antiviral production capabilities to produce
hundreds of thousands of doses in less than six months in 2005-2006, when their governments were
preparing for a potential H5N1 pandemic. This capacity for rapid scale-up of production suggests that the
actual time needed for establishment of a new production line may be much less than five years. As with
influenza vaccines, a general lack of domestic demand for influenza antivirals appears to be keeping
globalization of GoF benefits related to the development of novel therapeutics in check.
The U.S. demonstrated its willingness to donate antivirals during the 2009 H1N1 pandemic. However,
problems of timeliness of supply compounded issues of suboptimal use in-country. The WHO Pandemic
Influenza Preparedness Framework (developed in 2011) addresses these shortcomings but remains
untested.
1920 (2015g) Interview with US government official involved in public health preparedness and response decision-making for
influenza outbreaks.
1921 Dale Fisher et al. “Pandemic response lessons from influenza H1N1 2009 in Asia," Respirology 16 (2011): p. 879,
<http://onlinelibrary.wiley.com/doi/10.1111/j.1440-1843.2011.02003.x/abstract>.
1922 Ibid.

15.9.5 Potential Benefit 3- Benefits to pandemic preparedness planning
This section assesses the globalization of GoF benefits that inform pandemic preparedness planning,
which includes two benefits. First, the demonstration that avian influenza viruses can evolve the capacity
for more efficient transmission in mammals may, in and of itself, stimulate interest and investment in
pandemic preparedness initiatives. Second, molecular markers for phenotypic properties of concern (e.g.
virulence, transmissibility, mammalian adaptation, and antiviral resistance), which are discovered and
validated using GoF approaches, inform pandemic risk assessments that guide prioritization of resources
for pandemic preparedness activities. The first benefit derives from GoF research that enhances the
transmissibility of influenza viruses in mammals, and the second derives from GoF research that enhances
the infectivity or transmissibility of influenza viruses in mammals, that enhances the virulence of
influenza viruses, and that leads to evasion of influenza viruses from therapeutics. For a detailed analysis
of these GoF benefits, refer to individual benefit sections for each GoF phenotype.
Formal and informal pandemic risk assessments inform the extent to which governments invest in
pandemic preparedness and response activities as well as how those resources are directed, given that
many zoonotic influenza strains pose potential risks to human populations. These activities include
enhanced surveillance and implementation of interventions at the animal-human interface to mitigate risks
of disease spillover into human populations, in order to bolster prevention and early detection capabilities,
as well as development of pre-pandemic vaccines.
The extent to which GoF benefits to pandemic risk assessments will globalize depends on several factors:
• Whether and how information gleaned from GoF studies influences risk assessments and
decision-making about pandemic preparedness activities in developing countries in which high-
risk animal influenza viruses are currently circulating,
• Whether those countries have the ability to successfully implement community-level

interventions that mitigate the risk of disease spillover into human populations and that bolster
their capacity for early detection of potential spillover events, or
• Whether those countries have the capacity to produce pre-pandemic vaccines in-country.
In this section, we evaluate current capabilities and challenges for each factor in turn.
15.9.5.1 Role of GoF Research in Pandemic Risk Assessments for Developing Countries
This section evaluates whether and how information gleaned from GoF studies influences risk
assessments and pandemic preparedness planning in developing countries in which high-risk animal
influenza viruses are currently circulating. Two types of GoF studies are considered: (1) ““proof of
principle” demonstrations that particular animal influenza viruses can acquire pandemic properties (e.g.
transmissibility) in the laboratory, and (2) studies that establish molecular markers for phenotypic
properties of concern (transmissibility, virulence, etc.).
Although “proof of principle” experiments that demonstrate that an avian virus (e.g. H5N1) can acquire
the capacity for more efficient transmission in mammals have had minimal impacts on USG initiatives
due to the already high investments in pandemic preparedness, these GoF results have relatively greater
impacts on preparedness efforts in developing countries. One international public health official stated

that the experimental demonstration that H5N1 could evolve the capacity for airborne transmission in
ferrets was of “great importance” in countries where H5N1 was circulating.1923,1924,1925 In response, some
countries mounted communications campaigns to engage with the public, public health personnel, and
health care workers about the risks associated with H5N1, in an effort to bolster their surveillance
capabilities. Thus, to date, these GoF experiments primarily benefit global rather than domestic
populations.
As discussed in detail in section 15.3.5.2, risk assessments of particular virus strains integrate several
different types of information that influence the pandemic potential of a virus, including information
about the transmissibility, virulence, and other properties of the virus, information about pre-existing
immunity and other properties of the host population, and information about the circulation of the virus in
local animal populations and other ecological factors. Most developing countries in which animal
influenza viruses of concern (e.g. H5N1) are circulating are not capable of conducting ferret experiments
to evaluate the transmissibility and virulence of viruses, which contribute critical data to a pandemic risk
assessment.1926 As a result, those developing countries carry out risk assessments in conjunction with the
WHO (as well as the CDC and other laboratories in the GISRS as needed).1927 This collaborative
relationship is codified in the WHO’s Pandemic Influenza Preparedness Benefit Sharing System, which
states that WHO will seek to ensure that member states and the WHO Secretariat “provide pandemic
surveillance and risk assessment and early warning information and services to all countries.”1928 These
assessments are conducted with input from the Ministries of Health in a country of interest.1929 In addition
to conducting risk assessments when new viruses of concern emerge, the WHO regularly updates
previous risk assessments in light of new epidemiologic observations and new data that has been
generated since the previous assessment. Similar to risk assessments conducted by the USG, WHO risk
assessments consider the presence of molecular markers of mammalian adaptation, transmissibility, and
virulence, alongside virological data and in the context of environmental factors that play important roles
in the emergence of pandemic viruses.
Ultimately, the ability of a developing country to derive benefits from risk assessments informed by GoF
research will depend on the ability of the country to engage in responsive pandemic preparedness
activities. These include enhanced surveillance, implementation of community-level risk mitigation
measures, and pre-pandemic vaccine development.1930 The following sections assess the potential for
developing countries to put in place such “downstream” responses.
15.9.5.2 Capacity for responsive public health preparedness measures in foreign countries
Responsive capabilities are primarily relevant in countries in which zoonotic influenza viruses (or
influenza viruses with zoonotic potential) are currently circulating. As seen on the map below (Figure
15.4), most countries in the world have detected cases of zoonotic avian influenza in humans or in birds
1925 (2015f) Interview with international researcher or international public health official.
1927 Ibid.
access to vaccines and other benefits, p.15.
1930 C. Todd Davis et al., “Use of Highly Pathogenic Avian Influenza A(H5N1) Gain-Of-Function Studies for Molecular-Based
Surveillance and Pandemic Preparedness,” mBio 5, no. 6 (December 12, 2014) <http://mbio.asm.org/content/5/6/e02431-
14.full>.

within the last five years. Notably, a lack of detected cases may be due to poor detection and reporting
capabilities rather than the absence of avian influenza.1931
Figure 15.4. Countries that reported a detected case of zoonotic influenza in humans or birds within the last
five years.1932,1933,1934,1935
Many countries with AI detections are developing (low- or middle-income) countries, in particular most
countries with repeated detections (i.e. multiple years) and sustained outbreaks in domestic poultry
populations. Public health responses to zoonotic influenza outbreaks in developing countries are
particularly challenging due to limited resources for carrying out response activities and because of the
need for a strong and coordinated veterinary service – public health system. The veterinary services of
most developing countries greatly suffer from weak human organizational factors compounded by
resource constraints.1936 The lack of effective communication strategies for behavioral interventions that
will reduce risks of disease spillover, e.g. at poultry farms, live bird markets, etc., was also highlighted by
influenza researchers and public health experts as a major challenge.1937 Convincing the public to comply
with disruptive measures is difficult, and one expert noted the value of GoF research results in
1931 Tiaji Salaam-Blyther, “The 2009 Influenza Pandemic: U.S. Responses to Global Human Cases,” Congressional Research
Service, June 23, 2009, p. 11,
<https://www.acs.org/content/dam/acsorg/policy/acsonthehill/globalchallengesdiscussions/swineflu/crs-r40588-us-
responses.pdf>.
1932 H5N1, H5N6, H6N1, H7N2, H7N3, H7N7, H7N9, H9N2, H10N7, H10N8.
1933 World Health Organization (WHO), “Disease Outbreak News (DONs),” <http://www.who.int/csr/don/en/>.
1934 World Health Organization (WHO), “Monthly Risk Assessment Summary, Influenza at the Human-Animal Interface,”
<http://www.who.int/influenza/human_animal_interface/HAI_Risk_Assessment/en/>.
<http://empres-i.fao.org/eipws3g/>.
1936 J. Weaver et al., “Initial assessment of strategic plans for improving the performance of Veterinary Services in developing
countries: a review of OIE PVS Gap Analysis reports,” Rev. sci. tech. Off. int. Epiz. 32, no. 2 (2012): p. 631-645.
1937 (2015f) Interview with international researcher or international public health official.

strengthening the evidence basis for recommendations. Nevertheless, as the following cases will
highlight, developing countries can still mount a public health response in the face of zoonotic influenza
detections.
The following section presents comparative case studies of the public health response in Thailand,
Vietnam, and Laos to novel influenza infections in people. Thailand is an upper-middle-income economy,
while Vietnam and Laos are lower-middle-income economies.1938 All three countries had cases of highly
pathogenic H5N1 infections in humans and in domestic poultry starting in late 2003 and early 2004.1939
The purpose of these case studies is to highlight the successes and challenges associated with
implementing community-level interventions to respond to the presence of ‘risky’ influenza viruses
circulating in the native animal population.
The case studies demonstrate the overarching importance of a strong public health sector in being able to
benefit from pandemic risk assessments through implementation of prevention activities. The case of
Thailand, and to some degree Vietnam, showcase that a robust response to a significant public health risk
in middle-income countries is not impossible. The case of Laos is instructive in demonstrating that, for
countries with little initial public health surveillance, the benefits realized from implementing
community-level response measures downstream of a pandemic risk assessment can be marginal at best.
15.9.5.2.1 Case studies: Thailand, Vietnam, and Laos and the 2004 H5N1 outbreaks
The following comparative case studies showcase different public health responses to the 2004 H5N1
outbreaks in Vietnam, Thailand, and Laos. Vietnam was the first of the three countries to report cases in
humans, shortly followed by Thailand and eventually by Laos. Vietnam’s response was resource-
intensive but initially ad hoc, with mixed success. Thailand’s approach was all-encompassing, with good
success. Laos had virtually no response capabilities in 2004 and its response mostly focused on
establishing a national surveillance system.
Vietnam
Vietnam first reported H5N1 in poultry on January 8, 2004 and in humans on January 11, 2004.1940
Vietnam initially responded to the 2004 cases with ad hoc bird eradication and poultry movement
restrictions.1941 These measures proved ineffective, given the lack of nationwide surveillance and
coordinated response capabilities. In response, Vietnam launched a nation-wide surveillance effort in
2004, and that year, 2272 samples were collected from poultry, of which 515 tested positive for HPAI
H5N1.1942,1943 Detection sampling was greatly expanded in 2005, with 13,889 samples collected from
poultry, of which 1,317 tested positive.1944
1938 The World Bank, “Country and Lending Groups,” http://data.worldbank.org/about/country-and-lending-groups. Accessed
July 7, 2015.
1939 David A. Boltz et al., “H5N1 Influenza Viruses in Lao People’s Democratic Republic,” Emerging Infectious Diseases
(2006): p. 1593, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3290961/>.
1940 World Health Organization (WHO), “H5N1 avian influenza: Timeline of major events,” January 25, 2012, p.1,
<http://www.who.int/influenza/human_animal_interface/H5N1_avian_influenza_update.pdf>.
1941 Ricardo J. Soares Magalhaes, Dirk U. Pfeiffer, Joachim Otte, “Evaluating the control of HPAIV H5N1 in Vietnam: virus
transmission within infected flocks reported before and after vaccination,” BMC Vet Res. 6 (2010): p.1
<http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2898779/pdf/1746-6148-6-31.pdf>.
1942 Xiu-Feng Wan et al., “Evolution of Highly Pathogenic H5N1 Avian Influenza Viruses in Vietnam between 2001 and 2007,”
PloSOne 3, no. 10 (October 2008): 1-12, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2565130/pdf/pone.0003462.pdf>.
1943 See Table 1 in: Ibid.
1944 Ibid.

In addition to initiating enhanced surveillance, Vietnam’s response strategy heavily relied on
vaccination.1945 A mass vaccination program for poultry was launched in August 2005, and the
government announced in January 2006 that over 240 million birds had been vaccinated.1946 The
measures were not entirely successful. Specific challenges highlighted by Vietnamese practitioners
included a “lack of knowledge about viral behaviors, pathogenicity, transmission mechanism, [and]
treatment,” problems with recognition and reporting systems, insufficient collaboration between human
and animal health sectors, and a general lack of resources to implement “active surveillance and
research.”1947 Today, H5N1 is considered endemic in poultry in Vietnam, and sporadic cases of human
infection with H5N1 continue to be reported by Vietnam.1948
Thailand
Thailand was hard-hit by the emergence of H5N1 in-country, and suffered several fatalities.1949 Mass die-
offs at poultry farms in central and northern Thailand were noted starting in late 2003.1950 Through mid-
December 2003 to early 2004, neighboring countries such as China, Vietnam, Japan, and South Korea
reported H5N1 outbreaks.1951 In response to these reports, the Thai government deployed a human-case
surveillance program in December 2003, followed by a poultry surveillance program in mid-January
2004.1952 The human-focused effort identified 12 confirmed and 21 suspected influenza cases in country
through polymerase chain reaction and viral isolation of respiratory specimens taken from individuals
exhibiting symptoms similar to influenza.1953 The animal-focused effort focused on collecting cloacal
swabs from poultry farms throughout the country, and led to the official announcement of the discovery
of H5 HPAI in a chicken farm.1954 The Thai national reference laboratory announced the first cases in
both human and poultry on January 23, 2004.1955 Subsequent monitoring results retrospectively analyzing
poultry outbreaks in 144 villages made it clear that H5N1 had already been present in Thailand since the
end of 2003.1956
Although initially lambasted by the local press for its sluggish response, the Thai government put in place
aggressive measures in an attempt to eradicate the virus.1957 A systematic nation-wide survey to detect
1946 Ibid.
1947 Nguyen Tran Hien, “Avian Influenza In Vietnam: Situation and Lessons Learned,” p.17,
<http://www.fao.org/docs/eims/upload/250718/aj167e00.pdf>.
1948 Sharmi W. Thor et al., “Detection and Characterization of Clade 1 Reassortant H5N1 Viruses Isolated from Human Cases in
Vietnam during 2013,” PloS One 10, no. 8 (2015): p. 1-20,
<http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4526568/pdf/pone.0133867.pdf>.
11 (November 2005): p.1664-1672, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3367332/>.
1951 World Health Organization (WHO), “H5N1 avian influenza: Timeline of major events.”
11 (November 2005), <http://wwwnc.cdc.gov/eid/article/11/11/05-0608_article>.
1953 Tawee Chotpitayasunondh et al., “Human Disease from Influenza A (H5N1), Thailand, 2004,” Emerging Infectious
Diseases 11, no. 2 (February 2005): p. 201-209, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3320461/>.
1954 Thanawat Tiensin et al., “Highly Pathogenic Avian Influenza H5N1, Thailand, 2004.”
1955 Ibid.
1956 Ibid.
Countries’ Responses,” p.17.

infections was rolled out in October 2004.1958 Poultry and related products (feed, bedding, waste, manure)
were destroyed by veterinary authorities upon identification of the virus; over 40 million birds were
reported killed in a nation-wide campaign.1959,1960 Controls were placed on the movement of commercial
poultry and fighting cocks, enforced through mobile checkpoints set up in the most affected provinces.1961
Finally, oseltamivir tablets were produced and sold at subsidized prices, starting with 200,000 tablets
manufactured in February 2006.1962
As a result of these response measures, the last reported human case of avian influenza in Thailand was in
2006 and the last reported animal case of avian influenza was in 2008.1963,1964,1965
Laos
Laos first reported H5N1 in poultry one day following Vietnam’s announcement, in January 27, 2004.
The first reported human case was detected two years later, with an onset date of February 10, 2007.1966
Prior to the 2004 H5N1 cases in humans in neighboring Vietnam and Thailand, Laos had extremely
limited disease surveillance system. Select hospitals were operating an “early warning outbreak
recognition” system using phones and faxes, but the information was not shared with the country’s
Epidemiology Unit.1967 As a result, the unit was unable to implement pre-emptive measures to the 2004
outbreak.1968 The response appears to have been limited to the culling of some 98,000 birds at commercial
farms.1969
The country sought financial assistance abroad to implement country-wide public health reforms.1970,1971
Laos deployed a disease surveillance network starting in 2007, with 3 surveillance stations for influenza-
like-illnesses and one surveillance station for both influenza-like-illnesses and severe acute respiratory
illnesses.1972 The system was expanded in 2009, 2010, and 2011.1973 Electronic means replaced the phone-
and-fax communication system for the hospitals, while a phone-in system was rolled-out so that rural
areas could report cases to the national health authorities.1974 Since the 2004 H5N1 cases and up until
1958 David A. Boltz et al., “H5N1 Influenza Viruses in Lao People’s Democratic Republic,” p. 1593.
1959 Thanawat Tiensin et al., “Highly Pathogenic Avian Influenza H5N1, Thailand, 2004”;
1961 Ibid, p. 17-18.
1962 Ibid, p. 17.
1963 Ibid.
1964 OIE, World Animal Health Organization Database (WAHID), “Detailed Country(ies) disease incidence,”
<http://www.oie.int/wahis_2/public/wahid.php/Diseaseinformation/statusdetail>.
<http://empres-i.fao.org/eipws3g/>.
1966 World Health Organization (WHO), “H5N1 avian influenza: Timeline of major events,” January 25, 2012, p.16,
<http://www.who.int/influenza/human_animal_interface/H5N1_avian_influenza_update.pdf>.
1967 Bounlay Phommasack et al., “Capacity Building in Response to Pandemic Influenza Threats: Lao PDR Case Study,” Am. J.
Trop. Med. Hyg. 87, no. 6 (December 2012): p. 965-971, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3516098/>.
1968 Ibid.
Countries’ Responses,” CRS Report For Congress, August 24, 2006, p. 15-16,
1970 Ibid, p. 16.
1971 The World Bank, “Facility supports a coordinated and effective response to H5N1 in Lao PDR (English),”
<http://documents.worldbank.org/curated/en/2010/03/13160390/facility-supports-coordinated-effective-response-h5n1-lao-
pdr>.
1972 Bounlay Phommasack et al., “Capacity Building in Response to Pandemic Influenza Threats: Lao PDR Case Study,”
1973 Ibid.
1974 Ibid.

2011, 19 H5N1 outbreaks in poultry have been detected in Laos.1975 Since the rollout of the human-
monitoring system in 2007 and up until 2011, a total of 31 influenza-like illness outbreaks in humans
have been investigated; of these, 27 were confirmed as influenza cases.1976
Laos has either been relatively spared from H5N1 cases in humans, with the last human case reported to
WHO in 2007, or has had low case detection.1977 H5N1 has not been reported in-country since mid-2010,
and the recent emergence of H5N6 in poultry is due to a strain believed to have originated from China
rather than emerging from Laos.1978
These cases are instructive in determining whether a developing country could benefit from utilizing
pandemic risk assessments to prioritize response capabilities. Countries like Thailand, and to a lesser
extent Vietnam, have demonstrated the ability to mount public health responses in the event of a serious
health situation. Conversely, the downstream benefits of pandemic risk assessments are significantly
limited in developing countries that lack the means to implement prevention and enhanced surveillance
activities, such as the situation in Laos in 2004.
15.9.5.3 Capacity for pre-pandemic vaccine production
In addition to implementing community-level prevention and surveillance activities in response to a high-

risk pandemic risk assessment, developing countries could derive benefits from such assessments by
investing in pre-pandemic vaccine development and stockpiling. The influenza vaccine producers with
influenza vaccines on the market identified in developing countries (see Section 16.9.6) are all capable of
producing pandemic vaccine strains using CVVs obtained through the WHO framework, as explained in
Section 16.9.3.1 above. The map in Figure 15.5 shows an overlay of the developing countries with current
vaccine production capabilities and those in which zoonotic influenza viruses have been detected in bird
and/or human populations within the past five years. Only seven out of 28 developing countries with
zoonotic AI detections in humans or in bird populations over the past five years have the capacity to
produce vaccines in-country. This result highlights that a limited number of countries that may be at risk
of the emergence of a novel pandemic strain within their borders can benefit from pandemic risk
assessments through the development and stockpiling of pre-pandemic vaccines. Notably, WHO does not
stockpile pre-pandemic vaccines for use in developing countries, but is rather focused on ensuring real-
time access to pandemic vaccines during a pandemic as outlined in the Pandemic Influenza Preparedness
Framework.1979,1980
1975 Ibid.
1976 See Table 1 in: Ibid.
1977 The World Bank, “Disease Outbreak News- Lao People’s Democratic Republic,”
<http://www.who.int/csr/don/archive/country/lao/en/>.
1978 Frank Y. K. Wong et al., “Reassortant Highly Pathogenic Influenza A(H5N6) Virus in Laos,” Emerging Infectious Diseases
21, no. 3 (March 2015): p. 511-516.
1979 Immunizations SWGoIVa. Influenza A (H5N1) Vaccine Stockpile and Inter-Pandemic Vaccine Use Background Document.
http://www.who.int/immunization/sage/meetings/2013/november/SAGE_WG_H5vaccine_background_paper_16Oct2013_v
4.pdf. Last Update Accessed October 31, 2015.

Figure 15.5. Overlay of low- and middle-income countries with current or planned influenza vaccine
production capabilities and those that have reported AI detections in birds to OIE within the past five years.
Regions with AI detections are outlined in red. Countries (or regions) without vaccine production capabilities
are shaded in white, countries with current vaccine production capabilities are shaded in dark blue, and
countries with planned vaccine production lines are shaded in cyan.
15.9.5.4 Summary – Globalization of GoF benefits that inform pandemic risk assessments
capabilities to conduct ferret experiments evaluating the transmissibility and virulence of viruses, data
which critically inform pandemic risk assessments, risk assessments are carried out in collaboration with
the WHO and laboratory members of the GISRS (including the CDC). Similar to USG risk assessments,
these risk assessments incorporate information derived from GoF research, alongside epidemiologic and
virologic data, and environmental factors that influence the pandemic potential of the virus.

15.9.6 Information on influenza vaccine production in low- and middle-income countries
The following dataset lists vaccine producers outside of high-income countries with influenza vaccine
products on the market, with influenza vaccine R&D, or that formerly marketed influenza vaccines but
appear to be no longer actively producing vaccines. The following list was compiled through several data
sources:
• The 2011 WHO survey on global influenza production capacity, which identified 28
companies,1981
• The Developing Countries Vaccine Manufacturers Network (DCVMN) directories from 2014 and
2015, which list current and planned influenza vaccine manufacturer members;1982,1983,1984
• The International Federation of Pharmaceutical Manufacturers & Associations’ Influenza
Vaccine Supply Members list,1985
• The U.S. Department of Health and Human Services’ Influenza Vaccine International Capacity
Building Portfolio.1986
These were then supplemented by searches for potential manufacturers identified in the literature or in
news reports.1987
1981 Jeffrey Partridge, Marie Paule Kieny, “Global production capacity of seasonal influenza vaccine in 2011,” Vaccine 31, no. 5
(January 2013): p. 728-731, <http://www.sciencedirect.com/science/article/pii/S0264410X12015861>.
1982 While the DCVMN is a coordinating platform for vaccine producers in the developing world, certain DCVMN producers
are in countries that are currently classed by the World Bank as being High-Income countries (such as South Korea).
<http://www.dcvmn.org/IMG/pdf/directory.pdf>.
http://www.dcvmn.org/IMG/pdf/dcvmn_directory_2014.pdf. Accessed July 7, 2015;
1985 International Federation of Pharmaceutical Manufacturers & Associations (IFPMA), “IFPMA Influenza task force – IVS
Membership,” <http://www.ifpma.org/resources/influenza-vaccines/ifpma-influenza-task-force/ivs-membership.html>
1986 U.S. Department of Health & Human Services, “International Influenza Vaccine Capacity Building Portfolio,”
<https://www.medicalcountermeasures.gov/projectmaps/Who.aspx>.
1987 Jan Hendriks, Yan Liang, Bing Zeng, “China’s emerging vaccine industry,” Human Vaccines 6, no. 7 (2010): p. 602-607,
<http://www.tandfonline.com/doi/pdf/10.4161/hv.6.7.11933>.

In total, the products of 36 vaccine companies based outside of high-income countries were researched.
Of these, 18 were found to be actively producing influenza vaccines, 13 had R&D work for such a
product at various stages of completion, and 5 were apparently not currently producing or researching
influenza vaccines. The following table summarizes these findings. It is unlikely to be a complete listing,
given that few companies provide up-to-date information on vaccine R&D efforts at the pre-clinical trial
stage. In addition, some uncertainties remain in cases where a product was recently on the market (2014)
but does not currently appear on the company’s products page, and no news or business articles were
available to explain the absence. Indeed, the WHO relies on company survey data to gauge current and
near-future influenza vaccine production.1988
1988 World Health Organization, “Pandemic Influenza Preparedness (PIP) Framework 2013 Partnership Contribution
Questionnaire Final Results (30 May 2014),” May 30, 2014,
<http://www.who.int/influenza/pip/2013_PC_Final_Results_30May2014.pdf>.

Table 15.42. Current influenza vaccine production outside of high-income countries. Known current producers are emphasized in blue text
Vaccine Producers Current influenza Country World Bank Income Vaccine Network Source
vaccine producer? Ranking Association
1989,1990
Acerca de Birmex No / R&D ? Mexico Upper-middle DCVMN, BARDA/WHO
1991
Amson Vaccines & Pharma (pvt) Ltd Yes Pakistan Lower-middle
1992,1993
Beijing Minhai Biotechnology R&D China Upper-middle DCVMN
1994
Beijing Tiantan Biological Products Yes China Upper-middle DCVMN
1995
Bharat Biotech International Limited Yes India Lower-middle DCVMN
1996
Bio Farma Yes Indonesia Lower-middle DCVMN, BARDA
1997
Cadila Pharmaceuticals Limited Yes India Lower-middle DCVMN
1998
Cantacuzino Institute No Romania Upper-middle BARDA/WHO
1999,2000
Changchun BCHT Biotechnology R&D China Upper-middle DCVMN, BARDA/WHO
Changchun Changsheng Life Sciences 2001
Yes China Upper-middle
Limited
1989 No influenza product is explicitly listed in the company’s entry in the DCVMN 2015 directory, unlike what was stated in the DCVMN 2014 directory. See: Developing
Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p. 47-48.
1990 Érika Hernández, “Producirá México vacuna contra influenza,” [Mexico will produce an influenza vaccine], Reforma, July 13, 2015,
<http://www.reforma.com/aplicacioneslibre/preacceso/articulo/default.aspx?id=590386&urlredirect=http://www.reforma.com/aplicaciones/articulo/default.aspx?id=590386>.
1991 Amson Vaccines & Amson Pharma (PVT) LTD., “Product Profile,” <http://www.amson.org.pk/products.html>.
1992
Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p. 3-4.
1993 Minhai Biotechnology Co. Ltd., “Patents,” <http://en.biominhai.com/yfdt/&FrontComContent_list01-1369617220497ContId=56b25fd5-73e3-411e-8894-
ab9462fc265e&comContentId=56b25fd5-73e3-411e-8894-ab9462fc265e.html>.
1994 Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2015,” 2015, p. 5-6.
1998 “Institutul Cantacuzino nu face vaccin antigripal nici in sezonul 2015 - 2016, desi are autorizatii” [Cantacuzino Institute will not make flu vaccine in the 2015-2016 season,
despite having licenses], Ziare, May 21, 2015, http://www.ziare.com/social/spital/institutul-cantacuzino-nu-face-vaccin-antigripal-nici-in-sezonul-2015-2016-desi-are-
2000 PATH, “Signing of new Letter of Agreement between BCHT and PATH supports influenza vaccine development in China,”
<http://sites.path.org/vaccinedevelopment/files/2015/02/BCHTbulletin-on-agreement-with-PATH_020215_for-web-no-watermark.pdf>.
2001 Changsheng, “Influenza Split Vaccine,” <http://www.cs-vaccine.com/en/cp_page.asp?id=328>.

2002
China National Biotec Group Yes China Upper-middle DCVMN
2003
Dalian Aleph Biomedical No China Upper-middle
2004
Dalian Hissen Bio-pharm Yes China Upper-middle
2005
GPO R&D Thailand Upper-middle BARDA/WHO
2006
Hualan Biological Engineering Yes China Upper-middle IVS
2007
Incepta Vaccine Ltd Yes Bangladesh Lower-middle DCVMN
2008
Instituto Butantan Yes Brazil Upper-middle DCVMN, BARDA/WHO
Institute of Vaccines and Medical 2009,2010
R&D Vietnam Lower-middle DCVMN, BARDA
Biologicals (IVAC)
Institute of Medical Biology, Chinese 2011
R&D China Upper-middle DCVMN
Academy of Medical Sciences
2012
Jiangsu Ealong Biotech No China Upper-middle
2013
Panacea Biotec Limited Yes India Lower-middle DCVMN
2003 The company was acquired by Shanghai Fosun Pharmaceutical Co., Ltd. See: “Dalian Aleph Biomedical Co., Ltd.,” CMO CRO,
<http://www.cmocro.com/company/Dalian+Aleph+Biomedical+Co.,+Ltd./index.html>.
2004
Hissen, “产品中心: 流感病毒裂解疫苗（2014/2015）使用说明” [“Products: Influenza Virus Vaccine (2014/2015) Description”],
<http://www.hissen.com/products/View.aspx?id=185>.
2005 “Vaccine factory to restart construction,” Bangkok Post, December 11, 2014, <http://www.bangkokpost.com/lite/news/448902/vaccine-factory-to-restart-construction>.
2006 “Hualan is first influenza vaccine manufacturer in China to get WHO approval,” Vaccine News Daily, June 17, 2015, <http://vaccinenewsdaily.com/stories/510549688-
hualan-is-first-influenza-vaccine-manufacturer-in-china-to-get-who-approval>.
2010 Thanhniennews, “Affordable bird flu vaccine made in Vietnam passes first human trial,” Talk Vietnam, April 23, 2015, <http://www.talkvietnam.com/2015/04/affordable-
bird-flu-vaccine-made-in-vietnam-passes-first-human-trial/>.
2012 “It also halted production of the A/H1N1 flu vaccine in February when the quality permit expired, he said.” In: “Two Chinese Drug Makers Halt Production,” CRI English,
April 1, 2010, <http://english.cri.cn/6909/2010/04/01/1461s560685.htm>.

Production & Research Complex for 2014
R&D Iran Upper-middle DCVMN
Pasteur Institute of Iran
Queen Saovabha Memorial Institute, The 2015
R&D Thailand Upper-middle DCVMN
Thai Red Cross Society
2016
Razi Vaccine & Serum Research Institute R&D Iran Upper-middle DCVMN
Research Institute for Biological Safety 2017
Yes Kazakhstan Upper-middle BARDA/WHO
Problems (RIBSP)
2018
Serum Institute of India Ltd. Yes India Lower-middle DCVMN, BARDA/WHO
2019
Shanghai Fosun Pharmaceutical Yes China Upper-middle
Shenzhen Neptunus Interlong Biotech 2020,2021
Yes China Upper-middle
Co., Ltd.
2022
Sinovac Biotech Ltd. Yes China Upper-middle DCVMN
2023
The Biovac Institute Yes South Africa Upper-middle DCVMN, BARDA/WHO
The Government Pharmaceutical 2024
R&D Thailand Upper-middle DCVMN
Organization
2017 “Реестр отечественных поставщиков товаров фармацевтической и медицинской промышленности” [Register of domestic suppliers of goods Pharmaceutical and
Medical Industries], March 13, 2015, <http://arkalyk.kostanay.gov.kz/uploads/files/1d03853ecd302518be6a42de19ca184a.doc>.
2019 FosunPharma, “产业布局 > 核心产品 > 疫苗,” [“Industrial Distribution – Core Products – Vaccine”] <http://www.fosunpharma.com/products/ym>.
2020 Neptunus, “Company Profile,” <http://www.interlong.com/En/About/>.
2021 China Commodity Net, “Shenzhen Neptunus Interlong Bio-technique Co., Ltd. – Subunit Influenza Vaccine,” <http://ccne.mofcom.gov.cn/493005>.
2023 Developing Countries Vaccine Manufacturers Network, “DCVMN Directory 2014,” 2014, p.1-82, http://www.dcvmn.org/IMG/pdf/dcvmn_directory_2014.pdf. Accessed
July 7, 2015.

2025,2026
Tianjin Tasly Pharma No China Upper-middle
Torlak Institute (Institute of Virology, 2027
R&D Serbia Upper-middle BARDA/WHO
Vaccines and Sera)
2028,2029,
The Company for Vaccine and Biological
R&D Vietnam Lower-middle DCVMN, BARDA 2030
Production No.1 (VABIOTECH)
2031
VACSERA R&D Egypt Lower-middle DCVMN, BARDA/WHO
2032,2033
Walvax Biotechnology R&D China Upper-middle DCVMN
2025 “Tasly setting up flu vaccine base in Tianjin,” Research In China, July 31, 2007, <http://www.researchinchina.com/news/NewsInfo.aspx?Id=6428>.
2026 Tasly Holding Group Co. Ltd., “Products,” <http://www.tasly.com/en_web/Product_list2.aspx>.
2027 World Health Organization (WHO), Public Health Innovation and Intellectual Property (PHI), Department of Essential Medicines and Health Products (EMP), “Clinical
Research Organization (CRO) to support an Inactivated Influenza Vaccine Clinical Trial in Serbia,” Request for Proposal Bid Reference 2015/HIS/PHI/001, p. 1-33,
<http://www.who.int/phi/news/RFP_2015_HIS_PHI_001.pdf>.
2029 Juliet Bryant, “Influenza vaccine manufacturing in Viet Nam: Report on the APACI Satellite session,” One Health, 2015, <http://onehealth.org.vn/influenza-vaccine-
manufacturing-in-viet-namreport-on-the-apaci-satellite-session.new>.
2030 VABIOTECH, “Products – Vaccine,” <http://www.en.vabiotech.com.vn/index.php?option=com_content&view=article&id=88&Itemid=109&lang=en>.
2033 Walvax Biotechnology Co. Ltd., “产品宣传册,” [“Product Brochure"] <http://www.walvax.com/Model/6.aspx>.

15.10 List of Subject Matter Experts Interviewed for the Benefit Assessment
Table 16.43. List of Research Laboratories Visited

Principal Department Research Institution Research Focus
Investigator
Departments of Pediatrics and
Mark Denison Pathology, Microbiology, and Vanderbilt University Coronaviruses
Immunology
Ralph Baric Department of Epidemiology University of North Carolina Coronaviruses
St. Jude Children’s Research
Richard Webby Infectious Diseases Department Influenza viruses
Hospital
Charles Russell Infectious Diseases Department Influenza viruses
Hospital
Stacey Schultz- St. Jude Children’s Research
Infectious Diseases Department Influenza viruses
Cherry Hospital
Paul Thomas Immunology Department
Hospital
National Institute of Allergy and Coronaviruses and
Kanta Subbarao National Institutes of Health
Infectious Diseases influenza viruses
Yoshihiro Department of Pathobiological University of Wisconsin,
Influenza viruses
Kawaoka Sciences Madison
Adolfo Garcia-
Department of Microbiology Mt. Sinai Hospital Influenza viruses
Sastre
Melissa Uccelini Department of Microbiology Mt. Sinai Hospital Influenza viruses
Randy Albrecht Department of Microbiology Mt. Sinai Hospital Influenza viruses
Department of Medicine,
Walter Orenstein Emory University Influenza viruses
Division of Infectious Diseases
Department of Microbiology and
Anice Lowen Emory University Influenza viruses
Immunology
Department of Microbiology and
John Steel Emory University Influenza viruses
Immunology
In addition to the principle investigator listed above, postdoctoral fellows and other senior research staff,
graduate students, and/or laboratory technicians were interviewed during each site visit.

Table 16.44. List of Additional Stakeholders Interviewed
Name Title Institute Sector
Acting Director of the
Bright, Rick HHS/ASPR/BARDA Government
Influenza Division
Former Director of
CDC’s Influenza
Division, Former
Cox, Nancy CDC Government
Director of CDC’s
WHO Collaborating
Center for Influenza
Scientific Technical
HHS/ASPR/BARDA/Influenza
Donabedian, Armen Advisor and Chief, Late Government
Division
Stage Development
Associate Director for
Donis, Ruben Policy, Evaluation, and Influenza Division, CDC Government
Preparedness
Katz, Jackie Acting Deputy Director Influenza Division, CDC Government
Korch, George Senior Science Advisor HHS/ASPR Government
Lead, Health Economics
Meltzer, Martin CDC Government
and Modeling Unit
Senior Advisor to the
Morens, David National Institutes of Health Government
Director of NIAID
Robinson, Robin Director of BARDA HHS/ASPR/BARDA Government
Director, Pandemic and National Association of
Rose, Patrick Catastrophic County and City Health Government
Preparedness Officials
Advisor,
Office of the Assistant
Roth, Cathy Director-General, World Health Organization Government
Health Systems and
Innovation Cluster
Chief Veterinary
Vannieuwenhoven, Ty Officer, National HHS/ASPR/OEM Government
Disaster Medical System
Vice President and Chief
Pfizer Vaccines Research and
Dormitzer, Phil Scientific Officer: Viral Industry
Development
Vaccines
Senior Policy Analyst,
Woodrow Wilson
School of Public and
International Affairs; Princeton University; Inovio
Mahmoud, Adel Industry
Lecturer in Molecular Pharmaceuticals
Biology; Board of
Directors, Inovio
Pharmaceuticals

Plotkin, Stanley Executive Advisor Sanofi Pasteur Industry
Vice President of
Smith, Gale Novavax Industry
Vaccine Development
Associate Professor,
Department of University of Maryland School
Frieman, Matthew Coronavirus researcher
Microbiology and of Medicine
Immunology
Professor of
Perlman, Stanley University of Iowa Coronavirus researcher
Microbiology
Associate Professor,
Coronavirus and
Poon, Leo Faculty of Medicine, University of Hong Kong
influenza researcher
School of Public Health
Laboratory of Viral
Bennink, Jack Section Chief Influenza researcher
Diseases/NIAID/NIH
Animal Influenza Ecology and
Assistant Professor of
Epidemiology Research
Bowman, Andrew Veterinary Preventive Influenza researcher
Program, The Ohio State
Medicine
University
Laboratory of Viral
Brooke, Chris Postdoctoral Fellow Influenza researcher
Diseases/NIAID/NIH
Associate Professor of
Bucher, Doris Microbiology and New York Medical College Influenza researcher
Immunology
Erasmus University Medical
Fouchier, Ron Professor of Virology Influenza researcher
Center
National Wildlife Health
Hall, Jeffrey Research Virologist Influenza researcher
Center, USGS
University of North Carolina
Heise, Mark Professor of Genetics Influenza researcher
School of Medicine
Virology Laboratory, National
Ip, Hon Microbiologist Influenza researcher
Wildlife Health Center, USGS
Professor and Chair of
Microbiology; Professor
Palese, Peter Mount Sinai Hospital Influenza researcher
of Medicine, Infectious
Diseases
Royal Society
University Research
Fellow, Principal
Russell, Colin University of Cambridge Influenza researcher
Research Associate;
Department of
Veterinary Medicine

Institute of Virology and
Department of Medical
Schwemmle, Martin Professor of Virology Microbiology and Hygiene, Influenza researcher
Albert Ludwig University of
Freiburg
Professor of Infectious
Disease Informatics;
Director of WHO
Smith, Derek Collaborating Center for University of Cambridge Influenza researcher
Modeling, Evolution,
and Control of Emerging
Infectious Diseases
U.S. National Poultry Research
Swayne, David Center Director Center, Agricultural Research Influenza researcher
Service, USDA
Laboratory of Infectious
Taubenberger, Jeffery Section Chief Influenza researcher
Diseases/NIAID/NIH
Laboratory of Viral
Yewdell, Jonathan Section Chief Influenza researcher
Diseases/NIAID/NIH
Bloomberg
Casadevall, Arturo Johns Hopkins University Non-PPP research
Distinguished Professor
Professor of
Boston University School of
Microbiology; Director
Duprex, Paul Medicine; National Emerging Non-PPP research
of Cell and Tissue
Infectious Diseases Institute
Imaging
Professor of Theoretical
Fraser, Christophe Imperial College London Non-PPP research
Epidemiology
Professor and Associate
Chair, Department of University of Michigan
Imperiale, Michael Non-PPP research
Microbiology and Medical School
Immunology
Head of Special
Pathogens, Head of
Vector Design and
Public Health Agency of
Kobinger, Gary Immunotherapy; Special Non-PPP research
Canada
Pathogens Program;
National Microbiology
Laboratory
Professor of Harvard T.H. Chan School of
Lipsitch, Marc Non-PPP research
Epidemiology Public Health

Professor of
Microbiology and
Immunology; Co-
Relman, David Director of the Center Stanford University Non-PPP research
for International
Security and
Cooperation
Center for Health Security,
Chief Executive Officer Non-PPP research,
Inglesby, Tom University of Pittsburgh
and Director clinician
Medical Center

16 Appendix V: Findings Informing the Biosecurity Risk
Assessment
Purpose of the Biosecurity Risk Assessment (RA) 828
Methodology 829
16.2.1 Assessment of Malicious Actors 829
16.2.2 Historical Analysis 830
16.2.3 Identification of Malicious Actors, Acts, and Consequences 830
16.2.4 Defense Assessment of Governance of Defensive Measures 832
16.2.5 Interview Methodology 833
16.2.6 Semi-Quantitative Analysis of Plausible Threats 837
Definitions of Terms Used in the Threat Matrix 839

16.3.1 Malicious actors 839
16.3.2 Malicious Acts 840
16.3.3 Possible Consequences of Successful Act 843
Analysis of Malicious Actor Capabilities and Motivations 845

16.4.1 Lone Outsiders 845
16.4.2 Lone Insiders 846
16.4.3 Organized Criminals 847
16.4.4 Domestic Terrorists and Extremists 847
16.4.5 Transnational Terrorist Groups, including State-like Terrorist Groups 849
16.4.6 Foreign Intelligence Entities 851
Analysis of Historical Incidents 851

16.5.1 Assessment of Malicious Act Options for a Lone Outsider 851
16.5.2 Assessment of Malicious Act Options for a Lone Insider 855
16.5.3 Assessment of Malicious Act Options for Organized Criminals 859
16.5.4 Assessment of Malicious Act Options for Domestic Terrorists and Extremists 863
16.5.5 Assessment of Malicious Act Options for Transnational Terrorists, including State-like
Groups 868
16.5.6 Assessment of Malicious Acts Options for Foreign Intelligence Entities 872
Attacks against Laboratories 877
Biocrimes Committed by Individuals 881
Terrorist and Extremist Events Tied to Biological Warfare 888
Designated Foreign Terrorist Organizations and Biological Weapons 896
Detailed History of Known Terrorist Biological Weapons Programs 903

16.10.1 Al Qaeda 903
16.10.2 Jemaah Islamiyah 910
16.10.3 Aum Shinrikyo 910
16.10.4 Rajneesh Cult 912
16.10.5 RISE 914

Other Terrorist/Extremist Groups Linked in Some Fashion to Biological Weapons 915
The Islamic State of Iraq and the Levant (ISIL) 917
Biosafety and Biosecurity at U.S. Research Laboratories 920

16.13.1 Biosafety Levels, Select Agents, and Risk Assessments 921
Laws, Guidance, Policies, Practices, and International Agreements on Biosafety and Biosecurity928
16.14.1 Types of Governing Instruments 928
16.14.2 Laws, International Agreements, and Guidance Documents 932
Restriction of Fundamental Research, Dual Use Research of Concern and Recombinant DNA
Guidelines 969
16.15.1 National Security Decision Directive 189 969
16.15.2 Dual Use Life Sciences Research Concern 969
16.15.3 NIH Guidelines for Research Involving Recombinant DNA and Synthetic Nucleic Acid
Molecules 972
Analysis of Security Measures: Requirements, Implementation, and Gaps of Security Measures 973
16.16.1 Training 974
16.16.2 Personnel Reliability 976
16.16.3 Physical Security 980
16.16.4 Surveillance and Monitoring 984
16.16.5 Storage, Inventory, and Accountability Processes 987
16.16.6 Transfer, Shipment, and Chain-of-Custody Protocols 989
16.16.7 Emergency Response Protocols 993
16.16.8 Indirect Security Measure: Screening Framework Guidance for Providers of Synthetic
Double-Stranded DNA 995
16.16.9 Governance of Hazardous Chemicals 996
16.16.10 Gaps in Security Governance 997
16.16.11 Major Challenges 999
16.16.12 Knowledge Gaps 1000

16.1 Purpose of the Biosecurity Risk Assessment (RA)
The purpose of the semi-quantitative biosecurity RA is to provide information regarding the risk that
malicious actors would misuse the fruits of GoF research or intentionally cause an outbreak of engineered
strains. The risk of malicious actors acquiring pathogens for use as a weapon and the risk of accidental
infection caused by a malicious act (i.e., the release of infected animals from a laboratory) were
considered in the assessment. Given that biosecurity risk has two distinct components each with unique
vulnerabilities and consequences (malicious acts directed at a laboratory conducting GoF research and the
misuse of information generated by GoF research), our approach to these components is discussed
separately. For the biosecurity RA of malicious acts directed at a laboratory, the results are described as
semi-quantitative biosecurity risk information that can be understood in the context of, and relative to, the
quantitative biosafety risk information provided in the task described above. That is, this report will
highlight which threats pose as much risk as accidents and which types of biosecurity measures (security
systems or controls on information) are as critical to consider as the most important biosafety features.
influenza, SARS-CoV, or MERS-CoV virus is available. Therefore, to identify the types of malicious
actors and acts that may target a GoF laboratory, the analysis included an examination of historical
incidents involving life science laboratories and hospitals, evaluating the motivations and capabilities of
malicious actors, and determining if and how existing security measures affect the likelihood of success
of a malicious act.
Data collection included historical incidents within the past 25 years because information about incidents
from prior years is not necessarily available or high quality, and the governance and oversight of research
and life science laboratories differs from prior years. However, events beyond the 25 year mark were
included only if sufficient, quality information was available and they provided relevant information
about malicious actor interest, motivation, and/or capability, or malicious acts. This historical analysis
provides an evidence-based method to understand, in a qualitative way, the probability that an event
would occur and the type of resources these malicious actors bring to bear when targeting a laboratory.
Furthermore, all relevant information about laboratory biosafety and biosecurity security, whether
codified or not, was collected to ensure the development of a complete picture of the governance and
implementation of security and crossover safety measures at laboratories wherein GoF viruses are stored
or studied. Seasonal influenza and MERS-CoV are not select agents. However, MERS-CoV is studied in
biosafety level 3 (i.e., high containment) laboratories. Although the Biosafety in Microbiological and
Biomedical Laboratories (BMBL) recommends currently circulating seasonal influenza be studied at
biosafety level (BSL) 2 and not-currently circulating seasonal influenza be studied at higher containment,
the actual level of containment is determined by the institution’s biosafety risk assessment and
requirements from appropriate regulatory agencies. In general, research with seasonal influenza is
conducted at BSL 2, BSL2-Enhanced, ABSL3, or BSL3 depending on the virus, type of research, the
institutional risk assessment, and regulatory agency’s requirements.2034 SARS-CoV and H5N1 influenza
virus are Biological Select agents and Toxins (BSAT), and no GoF virus included in this process is a Tier
1 BSAT. Although influenza, SARS-CoV, and MERS-CoV are not Tier 1 BSAT, security governance of
Tier 1 pathogens was included because a Notice of Proposed Rule Making was issued about the elevation
of laboratory-generated, mammalian-transmissible H5N1 influenza virus to Tier 1 status.
All of the data collected about the potential threats and security governance were used to assess the
plausible threats facing laboratories that study or store GoF virus(s). These plausible threats serve as the
most probable events that could lead to a loss of containment from a biosecurity incident. Therefore, they
2034 Research Administrator Interviews.

were used to focus the quantitative analysis of local and widespread infections on those acts that are the
most plausible in today’s laboratory security environment.
16.2 Methodology
The GoF studies considered in this report remain restricted to research that achieves enhanced virus
production, enhanced pathogenicity, transmission in mammals, and/or evasion of natural immunity or
medical countermeasures in influenza, SARS -CoV, and MERS-CoV virus strains, consistent with the
framework proposed by NSABB.2035
The biosecurity RA is divided into two chapters. The first chapter evaluates the consequences of plausible
biosecurity risks posed by malicious actors and acts targeting laboratories in which GoF viruses are
studied or stored. The risks posed by the independent replication of published GoF research by malicious
actors is examined separately in the chapter on Biosecurity Risk of Information.
The assessment of “Malicious Acts Targeting a Laboratory” is grounded in knowledge about biosecurity
procedures at U.S. research institutions, biosecurity governance in the United States, and biological and
conventional threats facing U.S. research institutions; this assessment follows the methodology
section.2036 The malicious actors considered as part of this evaluation include a lone insider, a lone
outsider, organized criminals, domestic terrorists and extremists, transnational terrorists, and foreign
intelligence entities. Data was collected through analysis of open source material, which consisted of
reviews of government documents, the Bureau of Labor Statistics workplace incident database, peer-
reviewed journal articles, academic databases and working papers, mass media accounts, and public
documents on and imagery of select laboratories. This effort, conducted at the unclassified level, was
supplemented by interviews with biosafety and/or biosecurity officials and researchers at various
laboratories around the country, and with members of the law enforcement and intelligence communities.
The assessment was conducted in four stages: 1) identification of possible threats, including the type of
actor, type of deliberate security breach, and possible consequences of a successful breach (an assessment
of the “offense”); 2) identification of the layered security measures employed at U.S. research institutions
to mitigate malicious actor risk and any challenges associated with the implementation of those measures
(as assessment of the “defense”); 3) assessment of overall security risks using realistic scenarios that are
based on the information collected as part of steps 1 and 2; 4) evaluation of the potential for a plausible
threat to cause local or global outbreaks based on the epidemiological modeling of the consequences of
such threats; and 5) comparison of possible pandemic consequences of plausible threats involving GoF
viruses and non-GoF viruses.
The following subsections describe the methodology employed in conducting the offense assessment, the
defense assessment, and the interviews.
16.2.1 Assessment of Malicious Actors
The offense assessment identifies possible threats, including the type of actor, type of deliberate security
breach, and possible consequences associated with a successful breach.
2035 Gryphon Scientific (2015) “Conducting Risk & Benefit Analysis of Gain of Function Research: Initial Draft Workplan”.
2036 An overview of bioterrorism risk assessment methodologies can be found in: Bruce K. Hope, Sarah Elrod, “Risk
Assessment in Bioterrorism,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A. Zilinskas
(Hoboken: John Wiley & Sons, 2011), p. 543-547.

A threat matrix was created from the assessment of historical cases, current events, and discussions with
members of the intelligence community, who provided helpful context and background about potential
malicious actors and malicious acts to take into account. The threat matrix, which is described in the next
section, provides the basis for evaluating historical and possible threats associated with research
laboratories in the United States.
16.2.2 Historical Analysis
The first step in the offense assessment process involved a historical analysis of attacks against
laboratories (Appendix V: Sec 16.3), biocrimes committed by individuals (Appendix V: Sec 16.4), and
terrorist interest in biological warfare (Appendix V: Secs 16.5-16.9) In general, collection of historical
incidents was restricted to the 25-years between 1990 to 2015. Information about different incidents (i.e.,
biocrimes, laboratory attacks, or terrorist interest) varied in quality before 1990. In addition, laboratory
governance and security changed dramatically in the 1980s, 1990s, and 2000s suggesting little relevance
of older laboratory attacks or biocrime incidents. However, incidents that occurred before 1990 were
included if they provided an indication of actor motivation, interest and/or capability, or possible type of
act. Information about historical incidents involving biocrimes from 1975 to 2015, laboratory attacks
from 1990 to 2015 (two incidents included were from the 1980s), transnational terrorists from 1980 to
2015, and the domestic terrorist and extremists from the 1950s to 2015 was collected.2037 In total, eighty-
four generic malicious actor–malicious act pairings, and ninety-six malicious act–possible loss of
containment pairings, were considered.
Each of the malicious actor-malicious act pairings were then analyzed and grouped into sections by
malicious actor type in Appendix V, Sec 16.7. For each of the pairings, available historical cases were
identified in open source literature. For pairings with no historical precedent, the possibility of occurrence
of a case and its potential consequences was considered based on malicious actor motivations and
capability or similarity to historical incidents. This assessment was carried out by looking at the actor’s
potential motivation and capabilities to carry out the given act. Historical cases that shared some
similarities with a particular event were summarized, if sufficient information was available, as these
cases provided a snapshot of the motivation and capabilities of malicious actors in carrying out similar
events. Care was taken to identify certain cases where incidents may have occurred but, because of their
nature, may not be documented in open source reporting (such as potential covert entries conducted by
foreign intelligence entities). The findings from Sec. 7.4 were summarized in graphical fashion by filling-
in cells of the threat matrix to identify historical cases and hypotheticals.
A summary of possible malicious actors and acts is presented in Sec. 7.6.1 This section draws lessons
from the historical cases and hypotheticals considered in Sec. 7.4 and Appendix V, Sec.16.2-16.15.9. but
also considers significant changes that have occurred in malicious actor capabilities and/or opportunities
or motivations that could alter identified historical trends. In sum, this section presents a short-list of
malicious actors, acts, and consequences that deserve further attention. An analysis of plausible malicious
actor/malicious act combinations based on evaluating the offense and defensive measures together are
described in Sec. 7.6.
16.2.3 Identification of Malicious Actors, Acts, and Consequences
The malicious actors considered were:
• Lone outsider,
2037 The two incidents involving laboratory attacks in the 1980s were theft of infected animals.

• Lone insider,
• Organized criminals,
• Domestic terrorists and extremists,
• Transnational terrorists including state-like terrorist groups, and
• Foreign intelligence entities.
The attack vectors considered were:
• Armed assault,
• Bomb or arson,
• Physical covert entry,
• Cyber covert entry,
• Theft of pathogens,
• Theft of equipment or materials,
• Sabotage, elicitation of information,
• Subversion of employee,
• Insertion of operative,
• Self-infection, and
• Reckless acts.
The potential consequences of such acts that might lead to a disease outbreak include the release of
infected animals from/within a laboratory, the release of infected laboratory animals into the environment,
the cross-contamination of laboratory animals, the deliberate exposure of a laboratory worker, and the
removal of a pathogen sample from the laboratory. Three types of infections may result from outdoor
release of pathogen: the deliberate infection of wild or domestic animals, the deliberate infection of
laboratory workers, and/or the deliberate infection of members of the general public.
Other possible consequences include monetary or personal benefit by malicious actors. These
consequences are not included in the security risk assessment because they do not directly affect human
health. However, the consequence of personal benefit, rather than of exposure or release of an agent,
aligns more closely with acts carried out by some malicious actors.
This threat matrix was used to identify probable security scenarios that was used to model the pandemic
potential of an intentional release. Identifying these scenarios involved three distinct steps: 1) considering
historical examples; 2) extrapolating possible actor/act/consequence combinations based on actor
motivation and capability; and 3) analyzing the probable actor/act/consequences combinations (referred to
as plausible threats) after overlaying current defensive measures onto the threat matrix. Threats motivated
by financial or other personal gain were not assessed in the qualitative biosecurity risk assessment.
The parameters of the threat matrix is shown in Figure 16.1 (See Below).

Figure 16.1. Threat matrix of malicious actors, malicious acts, and possible consequences.
16.2.4 Defense Assessment of Governance of Defensive Measures
The assessment of defensive measures, including governance, identifies the layered security measures
employed at U.S. research institutions, and inconsistencies and/or challenges associated with the
implementation of those security measures. Laboratory defenses against malicious actors are derived from
both biosafety and biosecurity oriented policies and practices, since safety-oriented measures restrict the
operating environment and often provide security benefits. As such, safety measures that also provide
security benefits was considered as part of the overall defense assessment.
The defense assessment begins with an overview of the tiered, agent-specific, and experiment-specific
operating framework. This overview highlights what types of laboratory operating frameworks are
currently approved to work with the GoF pathogens of interest to this report. Included in the assessment
were the requirements and practices related to key aspects of malicious actor defense per the Statement of
Work for this effort, namely: personnel training; personnel reliability; physical security; surveillance and
monitoring; storage, inventory, and accountability processes; hazardous chemicals protocols; transfer,
shipment, and chain-of-custody protocols; and emergency response protocols.
Sources used include federal laws, federal regulations, Executive Orders, international and domestic
guidance documents, and official inspection reports released through the Freedom of Information Act
process. The U.S. Code and acts of Congress published in the Federal Register were used to retrieve
laws.2038 Federal regulations were retrieved for review through the Electronic Code of Federal
Regulations.2039 These sources were supplemented by peer-reviewed journal articles on implementation,
news articles, and information derived from interviews.
2038 U.S. Government Publishing Office http://www.gpo.gov/fdsys/browse/collection.action?collectionCode=PLAW

2039 U.S. Government Publishing Office, “Electronic Code of Federal Regulations” www.ecfr.gov.

16.2.5 Interview Methodology
16.2.5.1 Intelligence and Law Enforcement Officials
Gryphon Scientific conducted a series of interviews with Federal intelligence and law enforcement
officials to develop the final threat matrix and identify those threats that pose the greatest concern to U.S.
national security. The questions asked ranged from the types of actors that are thought to target biological
laboratories to the types of methods that could breach physical and cyber security measures. The
interview script is included in Appendix III: Sec 14.9.
At the For Official Use Only level, the interviews provided valuable insight into the various malicious
actor types, their motivations and capabilities, types of malicious acts, and types of consequences that
should be included in the threat matrix.
The final threat matrix was used to map: 1) possible threats based on motivation and capability derived
from information obtained from open source publications and the interviews conducted with intelligence
and federal law enforcement officials; and 2) historical examples primarily based only on open source
literature.
16.2.5.2 Research Institution Officials and Scientists
Gryphon Scientific reached out to seven research institutions that conduct gain-of-function studies as part
of the biosecurity risk assessment process. Research was paused at six of these institutions.2040 Scientists
and officials were interviewed from a total of six research institutions, as one institution opted not to
participate in the biosecurity interviews.
Also interviewed were principal investigators whose research was paused, students and staff in their
laboratories, directors of institutional environmental health and safety, biosafety officials, campus police,
and staff responsible for emergency response. For the five institutions that worked with select agents, the
responsible officials and alternate responsible officials of the Biological Select Agents and Toxins
Program were interviewed. At three institutions, the local FBI WMD Coordinators who serve as liaisons
between federal law enforcement and research institution officials were interviewed. At one institution,
the Vice President of Research, General Counsel, and Director of Human Resources spoke to project
staff. The interview script is included in Appendix III: Sec 14.12.
The majority of the biosecurity interviews were conducted concurrently with other team members
conducting the RBA, which enabled a deeper understanding of certain measures that cross over between
safety and security. Implementation of security and crossover safety measures at research institutions that
support gain-of-function research are included in the research governance section (Appendix III, Sec.
14.11).
The results of these interviews contributed to the development of security case scenarios by providing
greater understanding of how regulatory requirements are implemented at institutions that conduct GoF
influenza, MERS-CoV, and SARS-CoV research. These institutions do not represent all research
institutions that support infectious disease research, and specifically Biological Select Agents and Toxins
(BSAT) regulated research. Despite this shortcoming, the case scenarios developed to evaluate the
pandemic potential of malicious acts involving GoF research reflect accurately the safety and security
2040 Jocelyn Kaiser, “Moratorium on risky virology studies leaves work at 14 institutions in limbo,” Science Insider, November
17, 2014, http://news.sciencemag.org/biology/2014/11/moratorium-risky-virology-studies-leaves-work-14-institutions-
limbo.

conditions at the institution wherein this research is conducted. Therefore, the results are informative for
all institutions under the moratorium.
Information about the primary threats facing research institutions at which the interview was conducted
are included in the threat matrix (Sec. 7.4) and described briefly in the threat section (Appendix V, Sec.
16.2-16.9).
16.2.5.3 Interview Guide for the Biosecurity Risk Assessment
In-person interviews will be semi-structured to allow us to ask follow-up questions as necessary.
Interview Script for Intelligence and Law Enforcement Officials

Part 1: Capabilities and Motivations of Hostile Actors + Potential Hostile Act Modalities
• Overarching Question: What is the risk that may ensue based on the successful targeting of a
biolab facility in the U.S. on the part of a malicious actor (i.e. target attractiveness)?
• What are the various types of malicious actors that have posed or may pose a specific threat in
this area and what is their demonstrated or postulated motivation for targeting a biolab facility?
o What types of actors are known to have targeted laboratories or may find laboratories an
attractive target to:
 cause an intentional on-site release of an agent,
 cause facility disruption or destruction,
 acquire information, agent, or expertise for malicious purposes? and
 Can you provide specific examples of the above?
o What types of actors have joined or would be most likely to join laboratories to build their
own skills?
o Are any types of actors likely to acquire a strain from a laboratory but NOT use them (use
them in their own R&D programs or defensive programs).
o Is the distribution of these actors equal throughout the world or more concentrated in one or
more specific region(s), or one or more category(ies) of malicious actor?
• Are certain types of malicious actor threats and malicious act modalities more prevalent, more
likely and/or more concerning than others? Why or why not?
• Would a successful hostile act against a biolab facility achieve the stated or postulated objectives
of a given threat actor (see threat matrix)? More so than a hostile action against another type of
target?
• Would you recommend any adjustments to our draft threat matrix (provided in advance) based on
your knowledge and understanding of potential malicious actor threats to laboratory facilities? If
so, what are specific things we should improve or change?
• How might malicious actors target and take action against a laboratory to gain access to materials
or expertise relating to GOF research (i.e. tactics, techniques and procedures)?

• What specific capabilities are required to permit malicious actor access to or launch an attack
against a facility?
o Physical,
o Cyber, and
o Documentation
• In what ways have actors tried to gain access to facilities, materials and expertise relating to
advanced genetic engineering or, more specifically, GOF research?
• Can you recommend any additional studies, reports, analyses, real world case studies, etc. that
would be important for us to consider in better understanding actor capability, access and
motivation?
• Can you recommend anyone else who would be important for us to interview?
Information Risk Questions:
• Which actors, if any, are interested in the use of contagious agents in an attack? Does the
possibility that the US has a relatively robust public health system to mitigate an outbreak and
therefore many/most deaths may occur elsewhere figure into the calculus of these actors?
• Is influenza virus or MERS or SARS-CoV particularly of interest to any actor (compared to other
deadly, contagious agents)?
• Has any substate actor shown any interest in manipulating a biological agent to make it more
dangerous?
• Does any substate actor have the capability to manipulate a viral agent?
• How long is a substate actor willing/capable to work on developing an agent to execute an attack?
• Have any actors (state or substate) been known to insert operatives into a laboratory to gain
knowledge or skills in particular techniques in the life sciences for the purposes of developing a
weapon?
• Are any actors interested in agents that are countermeasure resistant?
• Does the publication in the scientific literature of various methods to modify a dangerous
pathogen increase state/substate actor interest in attaining a biological agent or modifying a
pathogen to make it more dangerous compared to the publication of just one route to modify a
pathogen? Or is a terrorist who is interested in modifying an agent going to seek out means to do
so from the literature, regardless of how many dual-use articles are published?
Interview Script for Environmental Health & Safety, Biosafety and Institutional Security Officials
Part 2: Gaps in Biosecurity policies, plans and implementation
• Overarching Question: What is the probability of an incident arising from shortcomings or

exploitation of vulnerabilities in the security of pathogens?

• Do the current biolab security policy/regulatory environment and the implementation of the
security requirements mandated therein adequately address the various types of potential
malicious actor threats?
• In your opinion, are there specific gaps in policy or regulation (including staff awareness and
training programs) or in the implementation thereof that represent an exploitable vulnerability?
(please address the areas listed below)
o Personnel reliability/security,
o Physical/electronic access control,
o Inventory/accountability processes,
o Pathogen storage protocols,
o Transfer, shipment, and chain-of-custody protocols,
o Surveillance and monitoring,
o Malicious actor detection,
o Incident reporting, and
o Emergency response protocols.
• What challenges do you face in implementing current federal regulations? How might these
challenges affect facility vulnerability (increase, decrease, or no change)?
• Are there state and local laws that increase the vulnerability related to unauthorized individuals
gaining access to information, counter federal regulations, or impose barriers to implementation
of federal regulations?
• What state and local laws decrease biolab facility vulnerability or otherwise support federal
regulations?
• Have you ever experienced a malicious actor threat to or act against your facility?
• Are representative security plans and training/awareness programs for high containment facilities
in alignment with governing policies/regulations and best practices, and, if so, do they adequately
address the threat? Are there specific gaps or concerns?
• In your opinion, if gaps exist in terms of policy or regulation or in the implementation thereof,
what are your recommendations to remedy them?
• In addition to policy and/or regulatory requirements, are there any best-practices for biolab
security (in use domestically or internationally) that you would recommend?
• To what degree does your institution interact with the local FBI WMD Coordinator to stay ahead
of potential threats and to inform them of potential problems?
• Can you recommend any additional studies, reports, analyses, real world case studies etc. that
would be important for us to consider?

Interview Script for Researchers
• Overarching Question: What is the probability of misuse or theft arising from authorized
laboratory staff?
• What processes/protocols exist in the laboratory and within the institution to prevent misuse of
research, theft of agent, or malicious use of an agent? Are there specific gaps or concerns?
• What types of biological security training do laboratory staff receive? Are there specific gaps or
concerns?
• What processes exist for researchers to report suspicious or unusual events or actions? Are there
specific gaps or concerns?
• What processes exist to interact with relevant institutional officials to identify and reduce security
risks associated with your research? Are there specific gaps or concerns?
• Do laboratory staff consider security (i.e., misuse of research or theft) a high priority concern?
• Can you recommend any additional studies, reports, analyses, real world case studies etc. that
would be important for us to consider?
16.2.6 Semi-Quantitative Analysis of Plausible Threats
From the assessment of the offense and the defense, “relatively high-risk” threat scenarios were created
that match the motivations and capabilities of the malicious actors, the malicious acts they may attempt
and, in light of the defenses arrayed against them, the outcomes these events are likely to have. Simply
put, combinations of actor and act were compared against the defense to choose a set that are most likely
to have a bad outcome out of all other possibilities. From this comparison, qualitative statements can be
made on the frequency that these acts are likely to be attempted (based on the historical record, the
motivation of malicious actors and the overall activity level of malicious actors) and how likely they are
to be successful.
16.2.6.1 Qualitative Analysis of Plausible Threats
Narrowing down the universe of possible threats (Sec. 7.4, Appendix V Sec. 16.2-16.9) to probable
threats associated with U.S. laboratories conducting GoF influenza, SARS-CoV, and MERS-CoV
research involves systematic evaluation of the ability of implemented security measures to prevent
malicious actors from accessing laboratory materials, animals, or pathogens and carrying out malicious
acts at the laboratories. This analysis consists of two steps: 1) assessment of malicious actor intent,
capability, and ability to access high containment research laboratories given existing security measures
and in light of historical occurrences and expressed interest; and 2) evaluation of the likelihood of success
of malicious acts in the presence of existing security measures and of a successful act resulting in virus
escape (i.e., loss of containment). The qualitative assessment of plausible threats is based on an analysis
of historical examples and motivation/capability of malicious actors. This approach eliminates completely
implausible scenarios, such as the use of drones to deliver packages inside a laboratory without detection,
from the analysis.

Figure 16.2 depicts the process diagram for analyzing plausible threats based on offense and defensive
measures.
Malicious Actor Malicious Act Infectious

Disease
Consequences
Figure 16.2. The process diagram is the analytical framework used in the study to assess plausible threats
based on the offensive and defensive measures. Red indicated low or no risk. Green indicates risk.
The first step of this analysis involves assessing the malicious actor’s intent to develop and use biological
agents as weapons and/or to breach a research laboratory to acquire the pathogen, material, equipment or
animal; capability that a malicious actor could commit a malicious act, including acquisition of a
pathogen; and ability to gain access to a high containment, research laboratory and its contents, regardless
of whether the laboratory is the source of the agent or the target of an attack. In assessing the intent,
capability, and ability of the different malicious actors listed in the threat matrix (Appendix V Sec.16.1),
the relative success of an insider compared to an outsider was assessed. This assessment is based on
analysis of the historical incidents and the evaluation of malicious actor motivation s and capabilities.
An actor needs to have sufficient capability to commit an act. Capabilities includes specialized skill,
expertise, access to materials, and support. Capabilities of individuals who 1) subvert or elicit an insider,
2) are an outsider trying to commit a malicious act, and 3) an insider intent on commit malicious acts will
be evaluated separately. Defensive measures preventing or introducing barriers to capability are
incorporated in the analysis.
The second step involves assessing the likelihood that a particular malicious act could be carried out
successfully given currently implemented security measures and the likelihood that a successful act could
cause a virus escape (i.e., loss of containment). This analysis is based on federal requirements for security
of Biological Select Agents and Toxins and implementation of crossover biosafety and biosecurity
measures at the research institutions that project staff visited as part of this project. Because no systematic
analysis has been conducted to identify the state of safety and security measures at all high containment
facilities and BSAT facilities, our analysis does presume that the institutions visited do not represent all
institutions with high containment facilities. However, research was paused at five of the six institutions

project staff visited as part of the GoF deliberative process, which is roughly one-third of the 14
institutions that received “stop work” orders from the National Institutes of Health.
Plausible threats that could be faced by U.S. research institutions that conduct GoF research with
influenza, SARS-CoV, and MERS-CoV viruses were produced from this analysis. These threats were
grouped into three broad categories: 1) overt acts; 2) covert acts exposing members of the public; 3)
covert acts exposing laboratory workers. Overt acts involve incidents, such as bombs or active shooters
that would trigger emergency personnel to respond. Covert acts involve acts that are not carried out
openly and about which emergency personnel may not be aware. Covert acts are divided into those
exposing the public and those exposing laboratory workers to capture differences in health monitoring
and familiarity of the viruses in the research laboratory and the symptoms they cause in infected
individuals between each group of people. These categories of plausible threats will be analyzed using
epidemiological modeling as described for the Biosafety RA.
16.2.6.2 Semi-Quantitative Epidemiological Modeling of Security Risk Scenarios.
For each of these “relatively high-risk” scenarios that result in a loss of containment, the possible
outcomes were modelled by linking to results from the Quantitative Biosafety Risk Assessment
(described above). For example, if a malicious act results in the accidental or intentional release of an
aerosol, calculations performed in the Biosafety RA can help determine what the consequences of that
release would likely be for an aerosol that initially infected any number of people. Quantitative analysis
will be conducted using different numbers of people exposed: 1 infected individual, 10 or less infected
individuals, or greater than 10 infected individuals. Analysis of the plausible threats is qualitative while
the consequences of the threat can be quantitatively assessed based on the epidemiological models
developed for the biosafety risk assessment.
16.3 Definitions of Terms Used in the Threat Matrix
16.3.1 Malicious actors
The threat matrix includes seven actors: lone outsiders, lone insiders, organized criminals, domestic
terrorists and violent extremists, transnational terrorists, state-like transnational terrorists, and foreign
intelligence entities.
The U.S. Code of Federal Regulations defines terrorism as “the unlawful use of force and violence against
persons or property to intimidate or coerce a government, the civilian population, or any segment thereof,
in furtherance of political or social objectives.”2041,2042 This definition was used in our study to distinguish
terrorists from criminals.
16.3.1.1.1 Lone outsiders
Lone outsiders are unaffiliated individuals who have an interest in attacking or gaining access to research
facilities, materials, agents, experimental protocols, or results. They are not affiliated with any particular
group. However, they may aspire to become members or may simply act in independent support of an
existing group. They are not a student or employee of a research facility. The motivations and capabilities
of these actors vary greatly.
2041 Federal Bureau of Investigation (FBI), “Definitions of Terrorism in the U.S. Code,” https://www.fbi.gov/about-
us/investigate/terrorism/terrorism-definition. Accessed on July 17, 2015.
2042 Federal Bureau of Investigation (FBI), “Terrorism 2002-2005,” https://www.fbi.gov/stats-services/publications/terrorism-
2002-2005. Accessed on July 17, 2015.

16.3.1.1.2 Lone insiders
Lone insiders are unaffiliated individuals who work in a research facility and have interest in using
research materials or agents to cause harm. These actors are of particular interest because many have the
scientific training to manipulate biological agents. Various research facility support staff have access to
research materials and agents, but they do not necessarily have the scientific capabilities required to
conduct experimental protocols. Similar to the variability of capabilities, the motivations of these actors
varies greatly, ranging from ideological radicalization to emotionally-motivated behavior.
16.3.1.2 Organized criminals
Organized criminals are defined here using the FBI’s definition of organized crime: “any group having
some manner of a formalized structure and whose primary objective is to obtain money through illegal
activities.”2043
16.3.1.3 Domestic terrorists and violent extremists
Domestic terrorists and violent extremists include groups and their members who vandalize, attack, or
otherwise harm facilities and individuals to make a political statement, protest for a cause, or “correct” a
real or perceived wrong. These groups include violent extremists, such as some animal rights or eco-
radical groups, and cults that use violence to achieve their goals. Some individuals who are affiliated with
extremist organizations have tried to acquire biological agents from culture repositories.
16.3.1.4 Transnational terrorists, non-state actors
Transnational terrorists, non-state actors refer to non-state groups that operate across national borders and
have similar ideological and/or political interests. This category encompasses those actors that are
attempting to control and govern territory and use tactics that do not conform to international norms for
war and nation-building. A prominent example of such an actor is the Islamic State of Iraq and the Levant
(ISIL). These groups use violence to achieve their political goals and to attack other nations or groups that
they perceive as enemies.
U.S.-based individuals radicalized by transnational groups are considered within the scope of this entry.
16.3.1.5 Foreign intelligence entities
Foreign intelligence agencies are a branch of a foreign government or of its armed forces tasked with
conducting espionage against other countries.2044 The term “Foreign intelligence entities” refers to
individuals working for nation states to collect information about research efforts.
16.3.2 Malicious Acts
The threat matrix includes 12 different malicious acts, which are divided into four categories: 1) kinetic
attacks including armed assault and bombs or arson; 2) covert entry including physical entry or cyber
breach; 3) theft of materials or pathogens by an outsider or insider; and 4) acts involving insiders, which
2043 Federal Bureau of Investigation (FBI), “Glossary of Terms,” http://www.fbi.gov/about-

us/investigate/organizedcrime/glossary. Accessed on July 13, 2015.
2044 For related definitions, see for instance the old Foreign Intelligence Surveillance Act of 1978.
“50 U.S. Code Chapter 36- Foreign Intelligence Surveillance,” U.S. Code Title 50 Chapter 36, Subchapter I- Electronic
Surveillance (§§ 1801-1812), https://www.law.cornell.edu/uscode/text/50/1801. Accessed August 11, 2015.

includes sabotage, elicitation, subversion of employee, insertion of an operative, reckless intentional act,
and self-infection.
16.3.2.1 Kinetic attacks
Armed assault
An armed assault refers to the use of firearms to harm individuals who work at research facilities or
forcibly gain access to facilities.
Bombs or arson
Bombs refer to any type of explosive used to attack individuals or breach laboratories. These explosives
can be homemade from chemicals, military devices such as grenades, or commercial devices. Arson
refers to the deliberate starting of a fire at a facility. This fire can be caused by an incendiary device or
other explosives.
16.3.2.2 Covert entry
Physical entry
Covert physical entry of a facility refers to the physical access of a research facility or laboratory by an
unauthorized individual without detection.
Cyber breach2045
Cyber breach refers to the non-physical and unauthorized access of, and subsequent interaction with, a
computer or other electronic device linked in some manner to laboratory research. It includes hacking,
denial of service attacks, insertion of a computer viruses, and other computer-based breaches to access
facility engineering systems, facility or laboratory computers, or human resource information. These
breaches could disrupt operations, facilitate theft of information, or tamper with engineering controls.
Care is taken to distinguish between attacks against devices owned by the researchers themselves,
Internet-connected devices within a laboratory, and so-called “air-gapped” network(s) within the
laboratory that are isolated from the Internet. Unauthorized access can be gained remotely in different
ways, for instance through malicious webpages (ex. watering hole attacks), email attachments (ex. spear
phishing attacks), or the insertion of malicious programs on USBs or CD-ROMs subsequently inserted
into the target network. These breaches could be used to steal information, facilitate a break-in, or
potentially to tamper with engineering controls.
Cyber breaches are included in the matrix because malicious actors have attacked computer systems.
However, they will not be analyzed because they likely do not result in direct human health
consequences.
2045 Whether cyber breaches of a laboratory could directly lead to human health consequences through sabotage is outside the
scope of this assessment. The Department of Defense (DOD)’s Defense Science Board Task Force considered the potential
threat of cyber-sabotage in their May 2009 assessment of DoD laboratory security, and recommended that an in-depth study
be conducted to determine the potential cyber threat against U.S. laboratories. As discussed in their report, a proper
assessment of potential cyber-sabotage threats against a U.S. biological laboratory would necessitate full on-site access to a
U.S. laboratory to “identify actual or potential access” to its IT infrastructure. Office of the Under Secretary of Defense For
Acquisition, Technology, and Logistics, Defense Science Board, “Report of the Defense Science Board Task Force on
Department of Defense Biological Safety and Security Program,” May 2009, p. xii, 18-19, 41,
http://www.acq.osd.mil/dsb/reports/ADA499977.pdf. Accessed September 27, 2015.

16.3.2.3 Acts caused by either insiders or outsiders:
Theft of pathogen
Theft of pathogen refers to the unauthorized removal of a pathogen from long-term storage or from
experimental samples.
Theft of materials, equipment, or information
Theft of materials refers to the unauthorized removal of research reagents, chemicals, equipment,
experimental kits, research notes, information, or other items from a research laboratory or facility.
16.3.2.4 Acts involving insiders:
Sabotage
Sabotage refers to the deliberate destruction of laboratory equipment, experiments, stocks, or results.
Often, these acts are driven by personal gain, revenge, competition, or other personal motivations, rather
than a desire to cause a deliberate release of agent. However, certain acts of sabotage have the potential to
cause a loss of containment, regardless of the actual intent of the malicious actor. For example, mixing
animals from different experiments or mixing infected and uninfected animals could result in cross-
contamination of research animals.
Elicitation
Elicitation refers to the manipulation of an individual to gain information about the research and facility.
Desired information could involve research activities, research animal housing, research results, pathogen
storage locations and procedures, and facility and laboratory operations, procedures, and security
measures.
Subversion of an employee
Subversion of an employee refers to an actor actively working against an employee to gain physical
access to research facilities and agents, steal pathogens, or acquire information about the research or
facilities.
Insertion of an operative
Insertion of an operative refers to a member of an organization, group, or nation that joins a research
laboratory or facility as a student, employee, or authorized visitor. The individual is not known to be a
malicious actor or to be affiliated with a malicious organization by the institution they are infiltrating. An
operative can insert themselves to gain access to information, pathogens, or laboratory individuals and
facilities. This individual also may join a laboratory to build his or her skills in carrying out a particular
set of experiments, gain access to key scientists, or provide themselves with an opportunity to acquire
reagents, agent, or equipment.
Reckless intentional act
A reckless intentional act refers to a situation wherein a deliberate act accidentally results in loss of
containment. For example, the deliberate release of infected research animals by animal rights groups
would result in release of agent into the environment. In the discussion about historical incidents, theft of

animals is categorized as a reckless act. However, we separate theft of animals in the qualitative plausible
threat analysis.
Deliberate self-infection
Self-infection refers to an individual who deliberately infects himself/herself. This action does not
presuppose that the intention for infecting oneself is to infect others; self-infection may be done in an
attempt to commit suicide, cause self-harm, or conduct an unauthorized experiment using him/herself.
16.3.3 Possible Consequences of Successful Act
The threat matrix includes eight possible consequences resulting in loss of containment, which are
divided into two categories: 1) loss of containment resulting in release of infected animals from and
within laboratories, release of infected animals into the environment, cross-contamination of laboratory
animals, exposure of laboratory workers, and removal of a pathogen from the laboratory; and 2) outdoor
release of pathogen resulting in infection of wild or domestic animals, infection of other laboratory
workers, or infection of the public.
16.3.3.1 Loss of Primary containment:
Loss of Containment includes intentional or accidental release of infected research animals or agent from
the laboratory. It does not refer to outdoor release such as release of agent directly from a facility into the
environment as a liquid or aerosol.
Release of infected animals from and within laboratories
Release of infected animals from and within laboratories refers to the escape of research animals from
their housing, hoods, or other spaces within individual rooms or between rooms of the containment
laboratory. This consequence does not presuppose that the animal leaves the containment facility.
Release of infected animals into the environment
Release of infected animals into the environment refers to the facilitated escape of research animals
outside of the containment facility and into biosafety level (BSL) 1 or 2 laboratories. Release into BSL 1
or 2 laboratories may lead to release of the pathogen outside the building and into the environment.
Cross-contamination of laboratory animals
Cross-contamination of laboratory animals refers to the mixing of research animals from different
experiments or the mixing of infected research animals with uninfected research animals. Often, animals
in experiments are housed separately from other experimental animals or unused animals. Experimental
animals are housed in vivariums. Special facilities exist for housing animals infected with Biological
Select Agents and Toxins (BSAT).
Exposure of laboratory workers
Exposure of laboratory workers refers to at least one laboratory worker that might be infected by a
pathogen through a needle stick, tear in personal protective equipment, animal bites or scratches, or other
means of exposing a worker to a pathogen.

Removal of a pathogen from the laboratory
Removal of a pathogen from the laboratory refers to the physical removal of the pathogen from
experimental samples, infected animals, or stored inventory.
16.3.3.2 Outdoor release of pathogen:
Outdoor release of pathogen includes liquid or aerosol release of a pathogen into the environment or
neighboring community. The neighboring community includes laboratory workers who do not work with
the pathogen and the broader public. Examples of release into the environment include efflux of
aerosolized agent into the atmosphere caused by reversal of air handling systems or removal of HEPA
filters and release of liquid agent into the soil from sabotaged pipes or disposal measures.
Infection of wild or domestic animals
Infection of wild or domestic animals refers to infection of household animals, livestock, or wild animals
because of liquid or aerosol release of pathogen into the environment. This event could include release of
infected research animals into the environment. In addition, it does include environmental release of agent
from animal carcasses that have not been properly disposed of.
Infection of other laboratory workers
Infection of other laboratory workers refers to the infection of individuals who do not work directly with
the pathogen, but do work in the same facility or on the same research campus. Infection may occur
through aerosol from contaminated equipment, improper decontamination or fixation of samples, or
residue on clothing or other materials that came in direct contact with the agent.
Infection of the public
Infection of the public refers to a release in which at least one individual of the general public is infected
with a laboratory generated or adapted pathogen. Such infections could be caused by the deliberate
release of pathogens into the atmosphere, ground water, or soil in close proximity to neighborhoods,
commercial spaces, and parks and other outdoor spaces.
16.3.3.3 Personal Benefit
Other possible consequences include monetary or personal benefit by malicious actors. These
consequences are not included in the security risk assessment because they do not directly affect human
health. However, the consequence of personal benefit, rather than of exposure or release of an agent,
aligns more closely with acts carried out by some malicious actors.
Monetary benefit
Monetary gain refers to financial benefit from selling stolen items on the black market or developing a
commercial product using stolen samples or data.
Release of information for personal benefit
Release of information for personal gain refers to individual or group benefit including getting ahead of
competitors, revenge, financial gain, and reputational gain.

16.4 Analysis of Malicious Actor Capabilities and Motivations
This section provides an overview of the motivations and capabilities of the malicious actors considered.
16.4.1 Lone Outsiders
16.4.1.1 Motivations
Lone outsiders could be motivated to carry out a malicious act by ideology, by personal grievances, for
personal gain, or by a combination thereof. A malicious act may also be the work of a disturbed
individual outsider, and, therefore, lack a rational, pre-mediated motive.
Examples of potential ideological opposition leading to a lone outsider carrying out a malicious act
against a laboratory include: jihadist ideology, radical animal rights beliefs, radical environmental beliefs,
radical anti-genetic engineering or anti-modernity ideologies, or anarchism, more broadly. Personal
grievances could be directed against one or more individuals working at the laboratory, for instance an
individual that rejected the lone outsider’s candidacy to work at the laboratory, or spurned a romantic
approach. Personal grievances could also be generated over the laboratory’s presence in the locale where
the individual lives, generating a radical “not-in-my-backyard” response. Finally, the use of illegal drugs,
the misuse of legal drugs, or mental health issues may lead to an irrational, unpremeditated, malicious
acts.
Since the potential motives behind a lone outsider act are extremely varied, the willingness of a lone
outsider to risk death or capture in an attack is unpredictable. Similarly, a lone outsider’s willingness to
make ill or kill is highly unpredictable.
16.4.1.2 Capabilities
Lone outsiders can have, and have exhibited, a wide range of capabilities. Whilst rare, a few malicious
lone outsiders such as Ted Kaczynski (the “Unabomber”), Eric Robert Rudolph, and Muharem
Kurbegovic (the “Alphabet bomber”), have been well-versed in bomb-making.2046,2047 Others, such as
Larry C. Ford (who was found to have possessed dangerous pathogens after his suicide), had professional
experience handling dangerous pathogens.2048 On average, lone outsiders are “often ineffectual,” in the
sense that their violent plots often fail to produce casualties.2049 When lone actors do successfully commit
direct violence, it is often through the use of firearms, and sometimes explosives. For instance, scholar
Ramon Spaaij’s study of 88 cases of successful lone actor terrorism in the 1968-2010 period showed that
43% of cases involved the use of firearms and 28% involved the use of explosives.2050 Lone outsiders
2046 Jeffrey D. Simon, “The Alphabet Bomber (1974),” Toxic Terror: Assessing Terrorist Use of Chemical and Biological
Weapons, ed. Jonathan Tucker (Cambridge: The MIT Press, 2001), 85-86.
2047 Federal Bureau of Investigation (FBI), “The Pursuit and Capture of Eric Rudolph: Part 1 of an Interview with FBI Exec
Chris Swecker,” May 2005, https://www.fbi.gov/news/stories/2005/may/swecker_051605.
2048 Jo Thomas, “California Doctor’s Suicide Leaves Many Troubling Mysteries Unsolved,” The New York Times, November 3,
2002, p. 1, http://www.nytimes.com/2002/11/03/us/california-doctor-s-suicide-leaves-many-troubling-mysteries-
unsolved.html?pagewanted=1.
2049 The conclusion was stated in general for lone actors, but is clearly applicable to the subset composed of lone outsiders.
Borum R, Fein R, Vossekuil B (2012) “Dimensional Approach to Analyzing Lone Offender Terrorism,” Aggression and
Violent Behavior 17, no. 5: 390.
2050 Restricting totals to U.S. cases sees an increase in the use of firearms.
Spaaij R (2012) Understanding Lone Wolf Terrorism Melbourne: Springer.
For a discussion of this study, also see: Paul Gill, Lone-Actor Terrorists: A Behavioral Analysis (Abingdon: Routledge,
2015), p. 16-17.

would most likely not have access to heavy weapons, such as rocket propelled grenade launchers, that
could be used to breach containment walls.2051
16.4.2 Lone Insiders
As with lone outsiders, lone insiders could be motivated to carry out a malicious act for ideological,
personal, or financial reasons, or carry out unpremeditated and irrational acts as a result of mental health
issues, legal drug abuse, or illegal drug use.2052 A lone insider carrying out a malicious act against a
laboratory may involve radicalization in sympathy to a jihadist ideology. Hypothetically, the possibility
exists that a sudden disgust with one’s research could lead one to turn to radical animal rights beliefs,
radical environmental beliefs, radical anti-genetic engineering or anti-modernity ideologies, or
anarchism— however, no such case was uncovered in open source reporting. Personal grievances against
one or more individuals working at the laboratory, such as those potentially developed as a result of poor
work relations, could drive an individual to engage in violent acts. Finally, the abuse of legal drugs, the
use of illegal drugs, sudden emotional trauma, or mental health issues may lead to an irrational,
unpremeditated, malicious act.
A lone insider’s willingness to risk death and willingness to kill is unpredictable.
Unlike lone outsiders, a lone insider potentially will be subject to periodic personnel surety screening,
ranging from simple checks for the use of illegal drugs to personnel reliability program and related
background and criminal history checks, depending on the nature of their work and the type of facility in
question. Although laboratory workers who work with Tier 1 Select Agents are screened and evaluated
periodically, the potential that a worker becomes radicalized, disgruntled, or disturbed after hiring and
between evaluations, cannot be discounted.2053 Moreover, a worker could manage to mask their radical
beliefs or mental health issues during initial screening.2054 Finally, a lone insider may not be part of the
universe of formally-screened personnel, if they are not working with Select Agents. As such, although
personnel reliability programs are a first barrier against lone insider malicious acts, they are not a panacea
for detection of this type of threat actor.2055
Given that a lone insider would be working at a laboratory, they are more likely than lone outsiders to
have legitimate access to pathogens, research materials, and dangerous waste. Lone insiders are also
much more likely to have training on the safe handling and movement of pathogens than a malicious lone
outsider. Moreover, since a lone insider has some degree of inside access, attacks can be more complex
than those launched by lone outsiders– for instance through the prepositioning of supplies within the
2051 Although a lone outsider is unlikely to have the resources to obtain heavy weapons, this possibility cannot be entirely
discounted, as such arms have fallen into the hands of organized groups in the U.S.in the past. For instance, law enforcement
officials uncovered an Army light antitank weapon during their raid of the compound maintained by the “The Covenant, the
Sword, and the Arm of the Lord” group. Jessica Eve Stern, “The Covenant, the Sword, and the Arm of the Lord (1985),”
Toxic Terror: Assessing Terrorist Use of Chemical and Biological Weapons, ed. Jonathan Tucker (Cambridge: The MIT
Press, 2001), p. 150.
2052 Biringer B, et al (2007) Security Risk Assessment and Management: a Professional Guide for Protecting Buildings and
Infrastructures Hoboken: John Wiley & Sons.
2053 Bunn M, Sagan S (2014) A Worst Practices Guide to Insider Threats: Lessons from Past Mistakes Cambridge: American
Academy of Arts and Sciences.
2054 AAAS, AAU, APLU, FBI, Bridging Science and Security for Biological Research: Personnel Security Programs, p. 7.
2055 Bunn M, Sagan S (2014) A Worst Practices Guide to Insider Threats: Lessons from Past Mistakes Cambridge: American
Academy of Arts and Sciences.

laboratory or through the exploitation of an internal safety system (such as a fire alarm). Finally, since a
lone insider has some degree of inside access, they are far more likely to succeed in stealing pathogens or
materials.
16.4.3 Organized Criminals
Organized criminals seek to make a profit, for instance through the theft of equipment for sale on the
black market, or through extortion and racketeering. Criminal organizations might consider stealing and
selling pathogens to terrorists, although no such historical examples are recorded in open sources. The
profitability and market value of pathogens on the black market is not known.
Study of crimes against high-profile targets, which can be used as a proxy for crimes against laboratories,
show that the capabilities deployed will be commensurate with the expected payoff.2056 That is, organized
criminals can hire or coerce highly qualified individuals and bring sophisticated equipment and weapons
to bear for high-value activities (such as coercing engineers to establish a dedicated radio network in
support of narcotics trafficking activities).2057,2058 The ability of an organized criminal group to recruit
individuals willing to risk death in an operation is proportional to its revenue. For instance, the wealthy
drug cartels can draw from numerous individuals willing to kill and risk death for the organization, but a
small-time thievery ring could not.2059,2060 The likelihood that organized criminals have advanced
scientific skill an access to pathogens is low.
16.4.4 Domestic Terrorists and Extremists
Domestic terrorists and extremists are motivated by a number of ideological doctrines and political
causes. The FBI categorizes malicious groups into far-left groups, and far-right groups, although a
domestic violent millenarian cult or home-grown Jihadi group unaffiliated with a transnational group
might arise in the future.2061 As demonstrated by the list of attacks against laboratories contained in
Appendix V: Sec 16.1, far-right groups have as of yet not attacked U.S. laboratories, whilst far-left groups
have routinely targeted U.S. laboratories. The willingness of group members to harm or kill individuals
also varies across these various types of terrorist and extremist groups, as explained below.
Far-left groups are currently mostly motivated by radical animal rights and radical ecological beliefs.
These ideologies directly justify attacks against laboratories, and these groups are responsible for several
breaches at U.S. labs. Animal rights extremist groups such as the Animal Liberation Front (ALF), and
some eco-radical groups like the Earth Liberation Front (ELF), explicitly forbid the killing of individuals,
2056 Reinstedt RN, Westbury J (1980) “Major Crimes as Analogs to Potential Threats to Nuclear Facilities and Programs,”
RAND Note N-1498-SL, prepared for Sandia laboratories.
2057 “Mexico navy smashes Zetas cartel communications network,” BBC News, September 8, 2011,
http://www.bbc.com/news/world-latin-america-14846866.
2058 Beckhusen R (2012) “Mexican Cartels Enslave Engineers to Build Radio Network,” Wired
http://www.wired.com/2012/11/zeta-radio/.
2059 See for example: Jo Tuckman, “Mexican officials: 43 killed in major offensive against drug cartel,” The Guardian, May 22,
2015, http://www.theguardian.com/world/2015/may/22/mexico-firefight-drug-cartel-region
2060 Cook C (2007) “Mexico’s Drug Cartels,” CRS Report for Congress http://ftp.fas.org/sgp/crs/row/RL34215.pdf.
2061 Ibid.

and members have so far adhered to this principle.2062 Therefore, while ALF and ELF are motivated to
target laboratories, they are not motivated to steal pathogens to harm others. This restraint is not universal
across all far-left groups, as exemplified by the now defunct, very small, eco-radical group R.I.S.E.,
sought to use pathogens to cause mass casualties (see Appendix V: Sec 16.5). Eco-radical groups who are
willing to kill scientists, using firearms and bombs, recently have emerged in Latin America and
Europe.2063,2064 In addition to the groups’ vigorous propaganda against synthetic biology, they have
targeted individuals working in the nanotechnology and nuclear sectors instead of high containment
laboratories.2065 In addition, they have not targeted U.S. institutions or researchers.2066
In general, far-right groups are currently motivated by anti-government beliefs, radical religious beliefs
associated with the Christian Identity movement, and racial supremacist notions.2067,2068,2069 With the
possible exception of federal laboratories, far-right group ideology currently does not promote attacks
against laboratories. No reports in open sources describe a far-right group as having targeted a U.S.
biological laboratory. However, a radicalized researcher at a laboratory may attempt to smuggle out
pathogens out for use against other targets. A select few far-right groups have shown some interest in
biological weapons through their perpetration of biological weapons hoaxes (summarized in Appendix V:
Sec 16.6), but no group has so far displayed any actual biological weapons capability. For instance, a
defunct group called the “Counter Holocaust Lobbyists of Hillel” sent agar and B. cereus in a petri dish to
a Jewish organization, and claimed the petri dish held B. anthracis, Y. pestis, or a chemical warfare agent
as part of an apparent hoax.2070 In general, far-right groups are likely to resort to violence and have carried
out mass killings.2071
Domestic jihadi groups not affiliated or commanded by transnational groups, or domestic violent
millenarian cults, have so far not emerged. Should such a group form, they are expected to favor mass
casualties, as with transnational terrorist groups attacking U.S. targets.2072
Capabilities vary across groups, mostly depending on the group’s motivation and end goals. Averaging
over a large number of past cases, when domestic and transnational terrorist groups commit acts of
2062 Ackerman G (2003) “Beyond Arson? A Threat Assessment of the Earth Liberation Front,” Terrorism and Political Violence
15, no. 4: 143-170.
2063 Phillips L (2012) “Anarchists attack science,” Nature (News) 485, no. 561 http://www.nature.com/news/anarchists-attack-
science-1.10729.
2064 Corral G (2011) “Stand up against the anti-technology terrorists,” Nature (News) 476, no. 373,
http://www.nature.com/news/2011/110822/full/476373a.html.
2065 Ibid.
2066 Ibid.
2067 Federal Bureau of Investigation (FBI), “Domestic Threat: White Supremacy Extremism,” May 22, 2012,
https://www.fbi.gov/news/stories/2012/may/extremism_052212/extremism_052212.
2068 Federal Bureau of Investigation (FBI), “The Terrorist Threat,” February 6, 2002, https://www.fbi.gov/news/testimony/the-
terrorist-threat-confronting-the-united-states.
2069 For a description of Christian Identity and far-right group ideologies, and an example of a far-right group, see: Jessica Eve
Stern, “The Covenant, the Sword, and the Arm of the Lord (1985),” p. 141-144.
2070 Carus W (1996) Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 110-111;
The B’nai B’rith International Jewish Monthly, Volume 111: 67, https://books.google.com/books?id=V--
3AAAAIAAJ&q=anthracis+Yersinia+Counter+Holocaust+Lobbyists+of+Hillel&dq=anthracis+Yersinia+Counter+Holocau
st+Lobbyists+of+Hillel&hl=en&sa=X&ved=0CC8Q6AEwA2oVChMI98TMwLKIxgIVOEaMCh0gNAC0.
2071 Shane S (2015) “Homegrown Extremists Tied to Deadlier Toll Than Jihadists in U.S. Since 9/11,” The New York Times,
http://www.nytimes.com/2015/06/25/us/tally-of-attacks-in-us-challenges-perceptions-of-top-terror-threat.html.
2072 For a discussion of the potential BW threat from domestic millenarian cults, see: Gary Ackerman, Markus Binder (for
START), “Anatomizing the Behavior of Chemical and Biological Non-State Adversaries,” PASCC Semi-Annual Workshop
on Strategic Stability and WMD, Washington, U.S.A., December 5, 2014, p. 11,
http://csis.org/files/attachments/141205_Ackerman_Slides_0.pdf.

violence, they frequently rely on the use of explosives (65-75% of cases).2073,2074 Raids against far-right
group safe houses have previously uncovered large amounts of weapons and explosives, demonstrating
that domestic terrorists and extremists have access to the requisite weapons and equipment to carry out
such attacks.2075 Yet, the review of malicious acts against U.S. laboratories presented in Appendix V: Sec
16.1 reveals no bombings or armed assaults launched against U.S. laboratories by far-right groups. Far-
left groups have relied on night-time break-ins, often followed by arson using incendiary devices
(Appendix V: Sec 16.1). These groups have mostly eschewed the use of firearms, although some
members have been known to own guns.2076,2077 In sum, our efforts have not identified any attacks on U.S.
laboratories by a domestic terrorist or extremist group that were intended to produce casualties. The
current non-use of firearms and explosives against U.S. laboratories is not due to a lack of capabilities,
but rather of motivation.
Several domestic extremist groups, both far-left and far-right, operate as decentralized cells. For example,
the ALF and ELF operate through small isolated cells, which then publicize actions through pro-group
outlets in the name of the overall organization.2078 This organizational structure reduces the capabilities
the group(s) can bring to bear against a single target, in return for greater resilience to law enforcement
actions.2079
16.4.5 Transnational Terrorist Groups, including State-like Terrorist Groups
The total number of terrorist groups targeting U.S. citizens is low compared to the overall number of
foreign terrorist groups currently in operation worldwide. One study found that a total of 395 terrorist
organizations were active in the 1998-2005 period, where “active” was defined as having committed at
least one attack in the given time period.2080 In contrast, only fifty-nine Designated Foreign Terrorist
Organizations are currently listed on the official U.S. list of active foreign terrorist organizations
maintained by the U.S. Department of State.2081
In general, modern transnational terrorist groups targeting the United States often are motivated by
extremist religious ideology.2082 These groups have tended to be violent Islamists, although one
transnational terrorist group studied in this section that targeted American alongside Japanese targets,
2073 Restricting totals to U.S. cases sees an increase in the use of firearms.
Spaaij R (2012) Understanding Lone Wolf Terrorism Melbourne: Springer.
2074 For a discussion of this study, also see: Paul Gill, Lone-Actor Terrorists: A Behavioral Analysis, p. 16-17.
2075 See the aforementioned example of an antitank gun seized in a raid: Jessica Eve Stern, “The Covenant, the Sword, and the
Arm of the Lord (1985),” p. 150.
2076 Federal Bureau of Investigation (FBI), “Most Wanted Terrorists: Daniel Andreas San Diego,”
https://www.fbi.gov/wanted/wanted_terrorists/daniel-andreas-san-diego/view.
2077 Moran H, Costanzo J (1997) “3 animal rights activists are back in court,” Deseret News,
http://www.deseretnews.com/article/603034/3-animal-rights-activists-are-back-in-court.html?pg=all.
2078 National Consortium for the Study of Terrorism and Responses to Terrorism (START), “Countering Eco-Terrorism in the
United States: The Case of Òperation Backfire’,” September 2012, p.12,
http://www.start.umd.edu/sites/default/files/files/publications/Countermeasures_OperationBackfire.pdf.
2079 This fact complicates law enforcement infiltration and monitoring of these groups. See: Ibid.
2080 Asal V, Ackerman G, Rethemeyer G (2012), “Connections Can Be Toxic: Terrorist Organizational Factors and the Pursuit
of CBRN Weapons,” Studies in Conflict & Terrorism 35: p. 230.
2081 U.S. Department of State, “Foreign Terrorist Organizations,” http://www.state.gov/j/ct/rls/other/des/123085.htm.
2082 The phenomenon that modern terrorism is conducted for radical religious beliefs is what terrorism scholar David C.
Rapoport called the fourth or religious wave of terrorism.
Rapoport D (2004) “The Four Waves of Modern Terrorism,” in Attacking Terrorism: Elements of a Grand Strategy, eds.
Audrey Cronin, J. Ludes Washington: Georgetown University Press.

Aum Shinrikyo, was a millenarian cult.2083,2084 Since transnational terrorist groups have turned toward
mass violence and are increasingly displaying a lack of strategic restraint, their members are likely to seek
mass casualties and are unlikely to negotiate during attacks.2085,2086,2087,2088,2089,2090
The transnational terrorist groups of concern in this section are engaged in very active propaganda and
recruitment efforts targeting Western citizens.2091 Propaganda documents by al Qaeda (central) in
particular have called on scientists and technicians to assist them in launching chemical or biological
weapons attacks.2092 Therefore, transnational terrorist groups may seek to recruit U.S. laboratory workers
to assist them in attacking a laboratory, or may inspire a U.S. laboratory worker to carry out a malicious
act in the name of the group.
The transnational terrorist groups considered here are well-funded, well-organized, well-armed, and
highly motivated groups. They are capable of orchestrating complex attacks and have suitable resources
to orchestrate long-term plots. They are able to recruit members willing to carry out suicide operations
and mass killings.2093 They may have a chemical or biological weapons program involving scientifically
trained individuals, as was the case with Aum Shinrikyo and al Qaeda (Appendix V: Sec 16.7).
The transnational terrorist groups considered as threats in this section have experience with explosives
and with a wide range of heavy weapons, and, therefore, are capable of breaching secure facilities if such
supplies and equipment could be brought to bear in the United States. At the high end of the capability
spectrum considered here is the rising Islamic State of Iraq and the Levant (ISIL) group, which controls
significant territory in Syria and Iraq and is considered a state-like group (Appendix V: Sec 16.9).2094 This
group has demonstrated its ability to recruit or coerce engineers and scientists both in Syria and Iraq and
in Western countries (Appendix V: Sec 16.9).
2083
Ibid.
2084 In particular, the group attempted to attack two U.S. naval bases in Japan. Richard Danzig, Marc Sageman, Terrance
Leighton, Lloyd Hough, Hidemi Yuki, Rui Kotani, Zachary M. Hosford, “Aum Shinrikyo: Insights into how terrorists
develop biological and chemical weapons, second edition,” Center for a New American Security, December 2012, p. 19,
2085 The degree to which modern groups have become less hierarchical and more prone to mass killings is, however, debated.
Hoffman B (2006) Inside Terrorism New York: Columbia University Press.
2086 Neumann P (2009) Old and New Terrorism Malden: Polity Press.
2087 Duyvesteyn I (2004) “How New is the New Terrorism?” Studies in Conflict & Terrorism 27, no. 5: 439-454.
2088 One early work on al Qaeda that remarked how globalized the group had become is:
Peter L. Bergen, Holy War, Inc.: Inside the Secret World of Osama Bin Laden (New York: Touchstone, 2001), p. 199-224.
2089 Other case studies restricted to al Qaeda include:
Brad McAllister, “al Qaeda and the Innovative Firm: Demythologizing the Network,” Studies in Conflict & Terrorism 27
(2004), p. 298-299, 303-306, 314-315;
2090 Jones C (2006) “Al-Qaeda’s Innovative Improvisers: Learning in a Diffuse Transnational Network,” Cambridge Review of
International Affairs 19, no. 4.
2091 Hill L, Deveau S, De Vynck G (2014) “Canadians from Calgary to Timmins heed ISIL’s tweets,” Bloomberg,
http://www.bloomberg.com/news/articles/2014-10-23/canadians-from-calgary-to-timmins-heed-islamic-state.
2092 Office of the Director of National Intelligence, Bin Laden’s Bookshelf,” http://www.dni.gov/index.php/resources/bin-laden-
bookshelf?start=1;
Retrieved under the “Now Declassified Material” folder: Abu-Salih Al Somali, “Terror Franchise: The Unstoppable
Assassin, TECHS Vital role for its success,” <http://www.dni.gov/files/documents/ubl/english/Terror%20Franchise.pdf>.
2093 Not all groups have the will and capability to carry out suicide attacks. However, suicide attacks have become more
common in general, and have been repeatedly employed against U.S. targets in particular. On the diffusion of suicide
attacks, see:
Michael C. Horowitz, “Nonstate Actors and the Diffusion of Innovations: The Case of Suicide Terrorism,” International
Organization 64, no. 1 (January 2010): p. 33-64. In particular, note the diffusion diagram, Figure 3, p. 59.
2094 Zachary Laub, Jonathan Masters, “The Islamic State,” Council on Foreign Relations Backgrounders, May 18, 2015,
http://www.cfr.org/iraq/islamic-state/p14811.

16.4.6 Foreign Intelligence Entities
Nation-states may want to collect information on U.S. research or actual samples of biological materials
through their foreign intelligence arms for a wide range of reasons. Such efforts may be carried out for
purely economic gain, as part of economic espionage efforts. They also may be driven by national
security matters, such as identifying U.S. biological agent countermeasure capabilities or aggressively
attempting to determine whether the the U.S. conducts biological weapons work. In other espionage
cases, foreign states may use the information collected to support their own domestic covert biological
weapons programs (such as the Soviet Union cases recounted in Appendix V Sec. 16.2.6).
Foreign countries that have targeted the United States in the past have nearly limitless capabilities to bring
to bear, including highly-trained scientists, access to research facilities and equipment, access to
pathogenic agents, significant financial resources, and sophisticated cyber-espionage tools. The limiting
factor regarding capabilities brought to bear against a U.S. laboratory will be the perceived payoff of the
malicious act considered, for instance the perceived value of the information slated to be stolen, and the
potential retaliatory consequences if discovered.
16.5 Analysis of Historical Incidents
16.5.1 Assessment of Malicious Act Options for a Lone Outsider
16.5.1.1 Armed assault
No cases of lone outsiders launching an armed assault against a U.S. laboratory have been uncovered.
However, the threat of active shooters on university campuses and other workplaces appears to be
increasing, which suggests that this type of attack should not be discounted.
16.5.1.2 Bombing or arson
Lone outsiders have set off bombs targeting individual biomedical scientists away from research facilities
as well as against health care centers, although apparently none against a research laboratory. Ted
Kaczynski (the “Unabomer”), acting alone, mailed bombs to several researchers, including to a geneticist
(Charles Epstein), in attacks motivated by an anti-modernity and anti-technology ideology.2095 Eric Robert
Rudolph launched a string of bombings in the U.S. against abortion clinics.2096 Therefore, a bombing or
arson attack by an outsider against a laboratory remains a possibility.
The chance of a bombing or arson leading to injury or death can be used as first approximation for the
chance of a severe bombing or arson attack that could lead to a loss of containment. This possibility does
undercount cases of bombing and arson not intending to cause loss of life either directly through the
bombing or indirectly through an outbreak (for instance bombings conducted after-hours, with sole intent
2095 Fox M (2011) “Charles Epstein, Leading Medical Geneticist Injured by Unabomber, Dies at 77,” The New York Times,
http://www.nytimes.com/2011/02/24/health/research/24epstein.html?_r=0. Accessed July 13, 2015.
2096 American Association for the Advancement of Science (AAAS), Association of American Universities (AAU), Association
of Public and Land-grant Universities (APLU), Federal Bureau of Investigation (FBI), Bridging Science and Security for
Biological Research: Personnel Security Programs, Meeting Report, Washington, United States, August 21-23, 2013, p. 41-
42, http://www.aaas.org/sites/default/files/reports/AAAS-APLU-AAU-
FBI%20report%20on%20personnel%20security%20070114.pdf. Accessed July 13, 2015.

to damage the lab); rather, the numbers can be used as a first approximation. Since the Bureau of Labor
Statistics began reporting bombing and arson as a separate injury category in 2011, only one such case
was reported in the 2011-2013 timeframe.2097,2098 The one reported bombing or arson event leading to
injury in the 2011-2013 period did not involve a laboratory (or a hospital).2099 Since this represents the
bombing or arson likelihood across all potential target types in the U.S., the chance that of a lone outsider
carrying out a successful bombing or arson against a laboratory is even lower, in part because high
containment laboratories are housed inside of facilities and lack direct access to outside windows or
doors. This suggests that a sizable bomb would have to be used to breach containment. Overall, although
a bombing or arson by a lone outsider is a technical possibility, the likelihood of its occurrence low– as
demonstrated by the lack of known cases and the probabilities cited for context.
16.5.1.3 Covert entry (physical) for theft of pathogen, material, animals, or information:
No lone outsider case of laboratory theft of pathogen or material has been found in open source reporting.
One uncharacterized case that might have been an attempt by an outsider to gain entry covertly involved
an attempted theft “targeted at the pathogen collection at the central reference laboratory for animal health
in Indonesia” that was “thwarted by security systems installed by the U.S. government,” but no further
information has been released that would allow one to identify the perpetrator type, motivation, and
capability.2100
Lone outsiders have surreptitiously obtained pathogens in other ways than theft from a laboratory, as
described below. No instances of a lone outsider breaking into a laboratory have been documented in
open source reporting, although petty theft (mostly targeting laptops) by outsiders has been anecdotally
described.
16.5.1.4 Covert entry (cyber) for theft of material, animals, or information
A covert cyber-entry may be used to steal research information, may facilitate a physical covert entry, or
may perhaps even be used to carry out sabotage. For the purposes of this section, the term “material”
encompasses digital information.
This section, focuses on the use of cyber-entry to facilitate physical covert entry through access to facility
information and security personnel lists and on the potential for cyber-sabotage through access to
engineering controls.
Three separate potential targets of cyber-breaches are considered:
1. The first potential target is an individual scientist’s personal computer, which is likely to hold
research material. Such a breach is statistically likely to occur. Data from 250,000 computers
from around the world running on a Windows operating system collected by the security firm
Kaspersky showed that approximately 5% of home computers were infected with active,
2097 Pre-2011 data on bombing and arson were merged with explosion and fire incidents.
Bureau of Labor Statistics, “Occupational Injuries/Illnesses and Fatal Injuries Profiles: Number of nonfatal occupational
injuries and illnesses […] Bombing, Arson,” 2011-2013. Data retrieved at http://data.bls.gov/gqt/InitialPage.
2098 Bureau of Labor Statistics, “Occupational Injuries/Illnesses and Fatal Injuries Profiles: Fatal occupational injuries […]
Bombing, arson,” 2011-2013. Data retrieved at: http://data.bls.gov/gqt/InitialPage. Accessed July 7, 2015.
2099 This yields a percent chance of a bombing or arson leading to injury or death well below 0.001% per year (alternatively
expressed as well below 0.005% over five years and well below 0.015% over 15 years).
2100 Committee on Prevention of Proliferation of Biological Weapons, Office for Central Europe and Eurasia, National Research
Council, The Biological Threat Reduction Program of the Department of Defense: From Foreign Assistance to Sustainable
Partnership (Washington: The National Academies Press, 2007), p.15, p.15 fn.4.

malicious software (malware).2101 In 2010, 40% of U.S. households surveyed by a study
conducted for Consumer Reports stated that they had had malware on their computer in the last
two years.2102 According to data collected by the market research firm GfK for Consumer Report
in 2013, a projected 58.2 million American adults had had at least one malware infection
affecting a home computer in 2013.2103 The overwhelming majority of these infections are not
tied to espionage activities, although numerous malicious programs are available even to lone
actors that would enable such activities.2104 The necessary technical skill needed to orchestrate an
attack can be relatively low, since certain malware with spying capabilities (spyware) are
designed to be user-friendly. For instance, the basic but latest version of the notorious spyware
ZeuS was reportedly available for about $700-1000 on the black market, with 24/7 technical
assistance offered.2105,2106
2. The second potential target is the laboratory’s Internet-connected computers, used by researchers
to conduct research at the facility. Although these systems may be more protected and monitored
than one’s home computer, the lab’s Internet-connected network will also have more use and
hence more risk of infection. A report by the U.S. Office of Management and Budget noted that
incidents targeting federal networks of a similar nature had increased from previous years,
reaching some 70,000 incidents in 2014.2107
3. The third is the laboratory’s internal computer network. These computers, in some cases, are “air-
gapped,” meaning that computers in the internal network and computers in networks that have
access to the Internet (and hence might be compromised from the outside) are not connected.2108
Breaching an air-gapped network would necessitate someone to connect a (likely unknowingly)
infected device, such as a USB stick, to a machine on the internal network. Moreover, exfiltrating
data out from the air-gapped network would be problematic, and require sophisticated
techniques.2109 For these reasons, malicious actors without physical access to the laboratory are
highly unlikely to be able to carry out attacks against air-gapped networks.
If access/security files and device engineering controls are kept on air-gapped networks, and if good
cyber-security practices are in place regarding access to the air-gapped systems, then the risks posed by
lone outsiders can be most likely limited to penetration of the first and second target networks. However,
since laboratories exhibited a wide range of device set-ups, laboratories likely vary in their level of cyber
security. That said, Biological Select Agents and Toxins laboratories are required to have information
2101 Kaspersky E (2013) “One in Twenty is the Sad Truth,” Kaspersky Lab https://eugene.kaspersky.com/2013/03/25/one-in-
twenty-is-the-sad-truth/. Accessed July 31, 2015.
2102 “Social insecurity: What millions of online users don’t know can’t hurt them,” Consumer Reports Magazine, June 2010,
http://www.consumerreports.org/cro/magazine-archive/2010/june/electronics-computers/social-
insecurity/overview/index.htm. Accessed July 31, 2015.
2103 Consumer Reports: 58.2 Million Americans Had a Malware Infection on Their Home PC Last Year,” Consumer Reports
Magazine, May 1, 2013, http://pressroom.consumerreports.org/pressroom/2013/05/my-entry.html. Accessed July 31, 2015.
2104 See for instance: “Trojan.Peskyspy- Listening in on your Conversations,” Symantec Official Blog, August 27, 2009,
http://www.symantec.com/connect/blogs/trojanpeskyspy-listening-your-conversations. Accessed July 31, 2015.
2105 Diane Bartz, “Analysis: Top Hacker “retires”; experts brace for his return,” Reuters, October 29, 2010,
http://www.reuters.com/article/2010/10/29/us-hackers-zeus-idUSTRE69S54Q20101029. Accessed July 31, 2015.
2106 Macdonald D, ed. Derek Manky, “Zeus: God of DIY Botnets,” FortiGuard Center,
http://www.fortiguard.com/legacy/analysis/zeusanalysis.html. Accessed July 31, 2015.
2107 Bennett C (2015) “Cyberattacks on federal government hit record high,” The Hill,
http://thehill.com/policy/cybersecurity/234601-cyberattacks-on-government-hit-record-high. Accessed July 31, 2015.
2108 Carrara B, Adams C (2014) “On Acoustic Covert Channels Between Air-Gapped Networks,” Foundations and Practice of
Security: 7th International Symposium, FPS 2014, Montreal, Canada, Revised Selected Papers, eds. Frédéric Cuppens,
Joaquin Garcia-Alfaro, Nur Zincir Heywood, Philip W. L. Fong (New York: Springer, 2015), p.4.
2109 Ibid.

security in place to prevent cyber breaches.2110,2111,2112,2113,2114,2115 Therefore, drawing any conclusions
about the potential for sabotage through tampering with engineering controls or grave facilitation of
physical access enabled by modification of security personnel lists is difficult.
Penetration of the first two types of target networks, namely personal computers of researchers and
Internet-connected laboratory computers, might prove valuable in facilitating unauthorized access to a
lone outsider. Access to these systems would reveal sensitive facility information and personal
information on facility personnel. Information such as the names and pictures of individuals, project
descriptions, personnel schedules, and the exact location of the laboratory within a broader facility could
potentially be gleaned from access to these target networks and could facilitate a physical access attempt.
Similarly, embarrassing information that might be gleaned from personal computers could be used to
subvert an employee through blackmail. However, whether this information could be gathered from open
sources is unclear. Interactions with researchers through social media, freely-available aerial imagery of
the facility area, descriptions of research projects on lab websites and researcher CVs, and research
publications could already provide a strong understanding of the targeted laboratory if combined.
Due to the significant problems surrounding attribution of cybercrimes, no reliable data exists on how
many cyber-breaches are the result of a lone malicious actor. In general terms, although there continue to
be lone actors who engage in cyber-penetration activities, it appears that hackers are working together
more often than in the past.2116
16.5.1.5 Sabotage
No instances of a lone outsider breaking into a laboratory have been described in open source reporting
(see above discussion under “covert entry”), and, therefore, no lone outsider case of laboratory sabotage
was found in open source reporting.
16.5.1.6 Elicitation of information
Although specific examples of laboratory workers being elicited by lone outsiders were not uncovered,
the aforementioned example of Eric Robert Rudolph shows that the prospect of a lone outsider eliciting
information from an employee to enhance their ability to hit a target cannot be discounted. According to a
summary provided by the Federal Bureau of Investigation (FBI), Rudolph “used flattery to befriend
young, female temporary employees, new administrative staff, and security guards at [abortion] clinics.
Through these techniques, he obtained information regarding security protocols, functions, and
scheduling in order to maximize the injurious effects of the attacks on the clinics.”2117
16.5.1.7 Insertion of operative
A lone outsider, by definition, does not have access to the laboratory.
2110 42 C.F.R. §73.11(c)(1).

2111 42 C.F.R. §73.11(c)(9).
2112 9 C.F.R. §121.11(c)(1).
2113 9 C.F.R. §121.11(c)(9).
2114 7 C.F.R. §331.11(c)(1).
2115 7 C.F.R. §331.11(c)(9).
2116 Zadig S, Tejay G (2012) “Emerging Cybercrime Trends: Legal, Ethical, and Practical Issues,” Investigating Cyber Law and
Cyber Ethics: Issues, Impacts and Practices, eds. Alfreda Dudley, James Braman, Giovanni Vincenti (Hershey: Information
Science Reference.
2117 AAAS, AAU, APLU, FBI, Bridging Science and Security for Biological Research: Personnel Security Programs, p. 41-42.

16.5.1.8 Reckless Act…
…Infection of outside animals (Wild or domestic): No cases where a pathogen was taken from a
laboratory by an individual insider and used to infect outside animals were found in open sources.
However, this incident type cannot be discarded, because, as noted above, lone outsiders have
obtained pathogens in the past to commit other crimes.
…Exposure of lab worker: This act would require a lone outsider to gain access to a laboratory’s
pathogen stocks or contained laboratory environment, which, as explained under the covert entry
segment above, has not been documented.
…Infection of public: The review of confirmed biocrimes from 1990 to 2015 carried out by
individuals with no laboratory access highlighted two cases of possession of a dangerous
pathogen (Larry C. Ford, Michael Just) and one case of attempted possession (Larry Wayne
Harris), alongside several cases of individuals infected with HIV who used their blood to
deliberately contaminate others. Overall, lone outsiders have demonstrated the willingness to use
pathogens to cause harm if they can obtain them. All but perhaps one case of pathogen possession
and attempted possession involved pathogens ordered from culture collections; how Larry C.
Ford came to possess dangerous pathogens is unknown. These cases occurred before the advent
of stringent dangerous pathogen regulations and hence should not be taken to represent the
current capability of a lone outsider to obtain dangerous pathogens.
16.5.1.9 Deliberate self-infection:
This act would require a lone outsider to gain access to a laboratory’s pathogen stocks or contained
laboratory environment, which, as explained under the covert entry segment above, has not been seen in
historical case reporting. Furthermore, the desire to commit suicide or harm others through self-infection
with a pathogen appears to be extremely low. A single case has been reported in the open literature. The
case did not involve a laboratory or a laboratory worker; rather, it involved a woman who attempted
suicide through HIV self-infection, probably with the help of an infected friend (see Sec 15.2).2118
16.5.1.10 Acts for which no specific examples were identified in open source reporting:
• Subversion of employee
• Reckless Act involving a point source release of a pathogen from a laboratory
• Reckless Act involving the release of infected laboratory animals from or within the laboratory
• Reckless Act involving infection of laboratory animals outside of containment.
• Reckless Act involving infection of environment
16.5.2 Assessment of Malicious Act Options for a Lone Insider
16.5.2.1 Armed assault:
The assessment presented under this type of act in the Lone Outsider section holds true with regards to
lone insiders. That is, while no cases of lone insiders launching an armed assault against a U.S. were
uncovered, such a malicious act could potentially occur in the future. Crimes, including murder, have
been committed against individuals at a lab by others from the same lab. The FBI identified one such
2118 This case is described in: W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 77.

recent case: the 2009 murder by asphyxiation of a graduate student named Annie Le by veterinary
technician Raymond Clark III.2119
16.5.2.2 Bombing or arson:
No lab bombing or acts of arson caused by a lone insider were found in open source reporting. Based on
Bureau of Labor Statistics data cited in the lone outsider section, there was only one reported bombing or
arson event leading to injury in the 2011-2013 period, and it did not involve a laboratory. In overall terms,
although a bombing or arson by a lone insider is a technical possibility, the likelihood of its occurrence is
low.
16.5.2.3 Covert entry (physical):
Disgruntled ex-researcher Mohsen Hosseinkhani provides the historical case underlying this scenario.
Although he had already been fired at the time of his crimes, he still held “insider” access to the
laboratory, since his access credentials had apparently not yet been revoked.2120,2121 He leveraged this
access to steal equipment and sabotage experiments.2122
16.5.2.4 Covert entry (cyber):
Please refer to the cyber covert entry in the Lone Outsider overview for a general overview of the
potential networks attacked and of the malicious acts that could be facilitated through successfully
penetrating these networks.
Unlike lone outsiders, a lone insider is likely to have physical access to computer systems housed in the
facility, and may even have physical access to the facility’s internal air-gapped network. This greater
access facilitates covert entry through the use of malware.
16.5.2.5 Theft of pathogen:
Four cases of lone insiders stealing pathogens from a laboratory were found in open source reporting.
Diane Thompson abused her position as a laboratory technician to steal Shigella dysenteriae from the
hospital laboratory to infect fellow workers; moreover, she had probably previously stolen a pathogen and
used it to infect her boyfriend.2123 Brian T. Stewart abused his position as a phlebotomist to steal HIV-
infected blood from his workplace, which he subsequently injected into his 11-year old son in an attempt
to kill him.2124 Finally, Richard J. Schmidt was a gastroenterologist that injected his former lover with
HIV and hepatitis using a contaminated hypodermic syringe obtained from work.2125
2120 Anemona Hartocollis, Al Baker, “Doctor Accused of Crimes Against Mice and Lab,” New York Times – City Room Blog,
December 2, 2011, http://cityroom.blogs.nytimes.com/2011/12/02/doctor-accused-of-crimes-against-mice-and-lab/.
2121 “Lab rat switcher jumps bail, flees to Iran,” Iran Times, http://iran-times.com/lab-rat-switcher-jumps-bail-flees-to-iran/.
2122 Ibid.
2123 Zilinskas R (2011) “Diane Thompson: A Case Study,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca
Katz, Raymond A. Zilinskas Hoboken: John Wiley & Sons.
2124 Carus W Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900.
2125 AAAS, AAU APLU, FBI, Bridging Science and Security for Biological Research: Personnel Security Programs, p. 38.

16.5.2.6 Theft of material or information:
Four cases of insiders stealing material from a laboratory were found in open source reporting. The
Harvard post-docs, Jiangyo Zhu and Kayako Kimbara, signed a statement admitting that they had stolen
research data, cell lines, and genetic material from the laboratory they were working in; the post-docs
were transitioning to another laboratory in the United States at the time and wished to use the materials in
their research.2126 Qingqiang Yin attempted to smuggle to China more than 250 vials, test tubes, and petri
dishes presumably containing bacteria and yeast that produced a valuable enzyme that he had stolen from
a Cornell laboratory he used to work at.2127 These were placed in his suitcase, and some were leaking, but
based on media descriptions of the incident it does not appear that the biological material involved was
pathogenic.2128 Yin presumably did so because he had not been re-hired by the laboratory and he was
attempting to obtain a position at a Chinese laboratory.2129 The case of Mohsen Hosseinkhani described
above is an example of an incident where a lone insider stole equipment and non-pathogen biological
products (stem cell cultures, antibodies) with commercial and research value.2130,2131 Hosseinkhani did so
for financial gain, but also out of a desire for revenge against having been fired.2132 Konan Michel Yao
stole and attempted to smuggle into the U.S. 22 vials containing DNA encoding Ebola genes taken from
his prior employer, the National Microbiology Laboratory (Canada).2133 He did so in an attempt to
transfer his prior research to his new employer.2134
16.5.2.7 Sabotage:
Instances of lone insiders sabotaging equipment and experiments have been found in open source
reporting. The review of attacks against laboratories and of biocrimes from 1990 to 2015 found two cases
of sabotage of equipment and/or experiments committed by lone insiders (Mohsen Hosseinkhani, Vipul
Bhrigu) and one case still in trial following a not guilty plea for reason of insanity (Ouyang Xiangyu).
Hosseinkhani and Bhrigu did not attempt to physically harm anyone; the two incidents were driven
instead by a desire for revenge and academic jealousy, respectively. These cases of sabotage did not
present a risk of release of a pathogen or of an infected animal. Ouyang Xiangyu was a graduate student
at Stanford University who allegedly sabotaged lab mates’ research by killing off their stem cells, and
then proceeded to attempt to poison lab mates and herself by putting paraformaldehyde in their water
bottles as well as her own.2135,2136 She pleaded not guilty due to insanity.2137
2126 Holland L (2006) “Couple Admits Cell Line Theft,” The Harvard Crimson,
http://www.thecrimson.com/article/2006/4/17/couple-admits-cell-line-theft-in/. Accessed September 10, 2015.
2127 Choi C (2002) “Lab theft conviction: Former Cornell researcher found guilty of stealing valuable enzymes,” The Scientist,
http://www.the-scientist.com/?articles.view/articleNo/21813/title/Lab-theft-conviction/. Accessed September 10, 2015.
2128 Ibid.
2129 Ibid.
2130 Anemona Hartocollis, Al Baker, “Doctor Accused of Crimes Against Mice and Lab.”
2131 “Lab rat switcher jumps bail, flees to Iran,” Iran Times.
2132 Ibid.
2134 Ibid.
2135 Sim M (2015) “A*Star scholarship holder Ouyang Xiangyu expelled from Stanford,” The Straits Times,
http://www.straitstimes.com/singapore/courts-crime/astar-scholarship-holder-ouyang-xiangyu-expelled-from-stanford.
Accessed September 10, 2015.
2136 “A*Star scholar charged for poisoning labmates’ drinks,” April 2, 2015, TR Emeritus,
https://web.archive.org/web/20150404222631/http://www.tremeritus.com/2015/04/02/astar-scholar-charged-for-poisoning-
labmates-drinks/. Accessed September 10, 2015.
2137 Ibid.

16.5.2.8 Elicitation of information:
Given widespread inter-lab information openness, as promoted by staff safety and security training and by
staff presentation of research, a Lone Insider likely would not need to elicit specific information not
already available to them. This situation may be different in select cases where a laboratory is connected
to a hospital or where a private industry laboratory conducts compartmentalized commercial work. In
such cases, a lone insider may wish to elicit information to allow them to move and transfer pathogens
and materials between such compartments (for instance, from the lab to the hospital) to cause some other
malicious act (such as stealing research or infecting the public).
16.5.2.9 Subversion of employee:
The assessment presented under this type of act in the Lone Outsider section holds true with regards to
lone insiders. That is, although no cases were identified of lone insiders subverting fellow lab workers,
analogous events have occurred before. For instance, a U.S. military police officer arrested in October
2011 for having attempted to sell military secrets had solicited his fellow soldiers for help in the scheme
before falling for an FBI sting operation.2138
16.5.2.10 Insertion of operative:
Insiders are, by definition, part of a laboratory.
16.5.2.11 Reckless Act…
…Cross-contamination of laboratory animals: Mohsen Hosseinkhani shuffled the name tags of

research animals to sabotage experiments. These animals were not contagious and, hence, there
were no risks of cross-contamination. No other relevant cases were identified in open source
reporting.
…Infection of outside animal (Wild or domestic): No recorded cases were found where an
individual insider took a pathogen from a laboratory and used it to infect outside animals.
However, this incident type cannot be discarded, because, as noted above, lone insiders have
taken pathogens out of laboratories to commit other crimes.
…Infection of lab worker: The review of confirmed biocrimes from 1990 to 2015 found one such
case (Diane Thompson).
…Infection of public: Four confirmed cases involving a pathogen obtained by an individual

insider to infect someone from the general public have been committed since 1990. These were:
the 2001 “Amerithrax” perpetrator(s), Richard J. Schmidt, Brian T. Stewart, and Diane
Thompson. 2139
2138 Federal Bureau of Investigation (FBI), “Insider Threat- Soldier Receives 16-Year Sentence for Attempted Espionage,” April
26, 2013, https://www.fbi.gov/news/stories/2013/april/soldier-receives-16-year-sentence-for-attempted-espionage/soldier-
receives-16-year-sentence-for-attempted-espionage. Accessed July 15, 2015.
2139 FBI strongly believes the perpetrator was a laboratory insider, Bruce Ivins. However, the latter committed suicide before the
case could be taken to trial. See: The United States Department of Justice, “Amerithrax Investigative Summary, Released
Pursuant to the Freedom of Information Act,” February 19, 2010, p. 6-11, 25-92,
http://www.justice.gov/archive/amerithrax/docs/amx-investigative-summary.pdf. Accessed July 14, 2015.

No cases of malicious or suicidal self-infection were found in the open literature. This excludes cases of
approved scientific self-experimentation. The desire to commit suicide or to harm others through self-
infection with a pathogen appears to be extremely low. Only one suicide attempt case has been reported in
the open literature, and the case did not involve a laboratory or a laboratory worker. Rather, it was an HIV
self-infection case involving a woman who attempted suicide, probably with the help of an infected
friend.2140
• Reckless Act involving a point source release of a pathogen from a laboratory

• Reckless Act involving infection of laboratory animals outside of containment.
• Reckless Act involving infection of environment
16.5.3 Assessment of Malicious Act Options for Organized Criminals
No cases were uncovered in open source reporting, and an armed assault does not match the perpetrator
type since such acts would not generate income and would place the criminals’ lives in danger. This
actor-act pairing can be discarded as unrealistic.
No cases were uncovered in open source reporting. However, the findings of a 1980 RAND analysis of
high-technology or high-value crimes are applicable here, since they describe robberies taking place
against secure compounds. The RAND study demonstrated that “perpetrators prefer to threaten violence
rather than use it,” and that violence was only threatened or used in robberies; in effect, there were no
uses of explosives as a means to gain access to a target.2141
Attacking a high-containment laboratory would expose the perpetrators to enormous risks. Arson in
particular would need to be carried out from the inside of the laboratory to cause significant financial
harm (and hence net gain to a competitor), which would require additional risk and sophistication.
Finally, such attacks lack credible monetary gain motivators. A bombing or arson would not generate
income, apart from the highly dubious scenario where a rival research group may stand to profit. In sum,
bombing and arson attacks by organized criminals against a laboratory are discarded as unrealistic
scenarios.
16.5.3.3 Covert entry (physical) for theft of pathogens or information:
No cases were uncovered in open source reporting, although organized criminals might potentially carry
out such operations in order to steal equipment, pathogens, or information. Organized criminals have
attempted to sell weapons-useable, non-biological, material stolen from high-security sites in the past. For
instance, criminals have been interdicted in selling a number of vials containing highly-enriched uranium
2140 This case is described in: W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 77.
2141 Reinstedt RN, Westbury J “Major Crimes as Analogs to Potential Threats to Nuclear Facilities and Programs”.

powder from a facility or facilities in Eastern Europe over the years.2142 However, theft of dangerous
pathogens for resale to terrorists has not been documented in open source documents. Terrorist groups
appear to have been unwilling to invest significant funds to support black market demand for pathogens.
For example, Al Qaeda’s BW program reportedly had a proposed start-up budget of $2000-4000 USD,
and one of their principal bioweaponeers routinely complained about a lack of money. 2143,2144 This low
profitability and market value for stolen pathogens suggests that organized criminal groups are unlikely to
target U.S. laboratories as a source for pathogens for sale on the black market.
Theft of equipment appears unprofitable. Indeed, one case of theft involving what a single individual
could carry out of a laboratory amounted to “only” $10,000 of losses.2145 Although additional individuals
may be able to increase their illegal profits by carrying out pieces of heavy equipment present at a
laboratory, these items could be stolen from less-protected venues instead.
Overall, laboratories are high-risk, low-reward targets from the perspective of an organized criminal
organization.
16.5.3.4 Covert entry (cyber) for theft of information:
Cyber-espionage by organized crime groups against researchers is uncommon. In their last publicly-
available Foreign Economic and Industrial Espionage report dated October 2011, the Office of the
National Counterintelligence Executive remarked that: “no evidence of involvement by independent
hackers in economic espionage has been found in intelligence or academic reporting to date, in large part
due to the absence of a profitable market for the resale of stolen information.”2146 At least one article
suggests that a possible organized crime group from Western Europe used computer hacking means to
steal information about studies on “biological warfare and nuclear physics” to sell to government
entities.2147 In addition, this group allegedly conducted more typical illegal activities, such as theft of bank
account and credit card information.2148
16.5.3.5 Sabotage:
The sabotage of a laboratory could conceivably lead to a relative gain for competitors, but the meager
profit margins and the high chance of detection make this an unlikely scenario. For instance, although a
competitor might derive profits from the elimination of a rival’s laboratory, a legal commercial purchase
of the rival firm or the hiring of a rival’s top scientist would be legal, probably cheaper, more likely to
succeed, and far less risky. This actor-act pairing can be discarded as unrealistic.
2142 The thorough nuclear forensic study of one such vial is described in: Kenton J. Moody, Patrick M. Grant, Ian D. Hutcheon,
Nuclear Forensic Analysis (Boca Raton: CRC Press, 2005), p. 401-419.
2143 Ibid.
2144 Pita R, Gunaratna R (2009) “Revisiting Al-Qaìda’s Anthrax Program,” CTC Sentinel Vol. 2 Issue 5,
https://www.ctc.usma.edu/posts/revisiting-al-qaida%E2%80%99s-anthrax-program. Accessed July 14, 2015.
2146 Office of the National Counterintelligence Executive, “Report to Congress on Foreign Economic Collection and Industrial
Espionage, 2009-2011,” October 2011, p. 10,
http://www.ncsc.gov/publications/reports/fecie_all/Foreign_Economic_Collection_2011.pdf. Accessed August 3, 2015.
2147 Shamah D (2014) “Israeli firm busts 13-year-long Europe hack attack,” Times of Israel,
http://www.timesofisrael.com/israeli-firm-busts-13-year-long-europe-hack-attack/. Accessed August 3, 2015.
2148 Ibid.

Elicitation of information has been employed by criminal groups before, although no specific cases
involving life science laboratories were found in open source literature.2149 Several sophisticated software
suites are available to private citizens that enable aggregating, visualizing, and finding patterns in large
amounts of public and non-public information. Criminal hacking groups are believed to be carrying out
such “dossier-building” activities to facilitate future hacks.2150 Information obtained through elicitation, in
particular through orchestrated social media interactions, can potentially be incorporated into these
dossiers.
No cases involving a criminal group subverting an employee at a life science laboratory were uncovered
in open source reporting. The aforementioned 1980 RAND analysis of high-technology or high-value
crimes demonstrated that the number of insiders that participate in a theft increases with the expected
illegal profit.2151 Coercion of employees by criminal groups has occurred before when the payoff was
believed to be very high; for instance, robbers have targeted the families of bank managers in an attempt
to coerce the latter to assist in particular robberies.2152 These gambits are complex, expensive, and
personnel-intensive operations for criminal groups to carry out, as at least two teams (the hostage takers
and the robbers) must work in coordination and must have conducted extensive reconnaissance to carry
out such an attempt. Since as noted above, laboratory thefts are likely not profitable, organized criminals
would have difficulty subverting insiders, and are unlikely to coerce employers.
16.5.3.8 Reckless act…
…Infection of outside animal (Wild or domestic): This scenario has occurred previously. In one
historical case, a pathogen was illicitly obtained by two criminals (Kevin T. Birch and James B.
Cahoon), and it was hypothesized that their end goal was to kill a race horse.2153 In another case,
New Zealand farmers admitted to having illegally introduced rabbit haemorrhagic disease for use
as a bio-control tool, after the use of the pathogen as a bio-control tool had been rejected by the
New Zealand Ministry of Agriculture and Forestry.2154 Although this last case stretches the
definition of “organized crime,” it is still a crime committed by a group of individuals for
financial gain, albeit an indirect one (by having more crops to sell, since less crops would be lost
to rabbits).
One potential additional case exists, but details remain insufficient to rule either way. Russian
officials told visiting U.S. National Research Council committee members in March 2007 that the
Russian Prosecutor’s office in Moscow had launched an investigation that year into “alleged
unsuccessful efforts to attack a large suburban chicken marketplace by introducing chicken
affected by avian influenza virus, which would cause the marketplace to close and business to
2149 Federal Bureau of Investigation (FBI), “Internet Social Networking Risks,” https://www.fbi.gov/about-
us/investigate/counterintelligence/internet-social-networking-risks. Accessed August 11, 2015.
2150 Robert Graham, “Because dossiers,” Errata Security, June 16, 2015, http://blog.erratasec.com/2015/06/because-
dossiers.html#.VbfWTPnZViY. Accessed August 11, 2015.
2151 Reinstedt RN, Westbury J (1980) “Major Crimes as Analogs to Potential Threats to Nuclear Facilities and Programs”..
2152 Ibid.
2153 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, Working Paper (Washington:
Center for Counterproliferation Research, National Defense University, February 2001 Revision), p. 101.
2154 Ibid.

shift to a competing marketplace.” 2155 Russian media reports on this event presented a different
perspective, specifically that the investigation was for “violation of veterinary rules which
negligently caused the spread of epizootic diseases or other serious consequences.” However, this
statement contradicts previous statements made by a prosecutor who told the Russian media that
he was not ruling out the possibility that the infected birds sold at the market had been “infected
intentionally shortly before the sale.”2156 The results and scope of this investigation are not
known.
…Exposure of lab worker: No cases were uncovered in open source reporting. No realistic
scenarios exist where such acts would both generate significant illegal profits and do so in a
manner that could not be conducted in an easier manner. This actor-act pairing can be discarded
as unrealistic.
…Infection of public: No cases have been uncovered in the open source literature on the use,
attempted use, acquisition, attempted acquisition, or development of pathogens as weapons
against individuals by organized crime groups.2157 The use of chemical poisons, including toxins,
by organized crime groups has been documented. For instance, Chinese and Russian contract
killers have reportedly used a toxin derived from the Gelsemium plant genus to poison their
victims.2158 Although information is scant, poisons are probably used by organized crime groups
because: they are comparatively easy to conceal and use against a target without endangering the
assassin; they are very likely to kill once introduced into the victim’s system; the delayed onset of
symptoms of some poisons provides time for the perpetrator to escape; and because the use of a
rare poison has a chance to be missed in an autopsy. These characteristics are not completely
shared with the pathogens investigated in GoF laboratories.
No cases were uncovered in open source reporting. Such acts do not match the perpetrator type, as no
realistic scenarios exist where such acts would generate significant illegal profits and do so in a manner
that could not be conducted in an easier manner.
• Insertion of an operative (the significant time and resources needed would go beyond most
organized crime groups’ resources),
• Reckless Act involving a point source release of a pathogen from a laboratory (these acts do not
match the perpetrator type),
(these acts do not match the perpetrator type),
2156 Михаил Алексеев [Mikhail Alekseyev], “Пернатая зараза [Fowl Infection],” Lenta.ru, February 19, 2007,
http://lenta.ru/articles/2007/02/19/flu1/. Accessed October 15, 2015.
2157 For instance, the following review article on organized crime’s multi-faceted challenges to public health did not raise such
scenarios: Lucy Reynolds, Martin McKee, “organised crime and the efforts to combat it: a concern for public health,”
Global Health 6 (2010): p. 21, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2996357/. Accessed July 13, 2015.
2158 Whitehead T (2015) “Fears Russian tycoon Alexander Perepilichnyy may have been poisoned with rare plant,” Telegraph,
http://www.telegraph.co.uk/news/uknews/11614054/Fears-Russian-tycoon-Alexander-Perepilichnyy-may-have-poisoned-
with-rare-plant.html. Accessed July 13, 2015.

• Reckless Act involving cross-contamination of laboratory animals (these acts do not match the
perpetrator type), and
• Reckless Act involving infection of laboratory animals outside of containment (these acts do not
match the perpetrator type).
16.5.4 Assessment of Malicious Act Options for Domestic Terrorists and Extremists
No such cases were found in open source reporting. A review of cases of attacks on laboratories (see Sec.
15.1) shows that laboratories have been frequent targets of animal rights extremists, principally by the
Animal Liberation Front (ALF) and sometimes by the Earth Liberation Front (ELF). As noted above, both
ALF and ELF doctrinal documents reject the killing of individuals.2159 ALF has never used firearms in
attacks. However, other eco-radical groups, such as the defunct U.S.-based R.I.S.E. cell (see Sec. 15.5)
and the Mexican-based eco-anarchist group Individuals Tending to Savagery, have been more inclined
towards violence.2160,2161 Therefore, although an armed assault against a laboratory by a domestic terrorist
or extremist group would be a novel occurrence, it remains a viable scenario.
Arson attacks are a trademark of the ALF, as well as of eco-radical groups like the ELF.2162 The review of
recent attacks (from 1989-2015) against laboratories (see Sec. 15.1) documents five arson attacks against
U.S. laboratories.
Although no bombings at U.S. biological laboratories were uncovered in open source reporting,
commercial buildings owned by biotechnology companies have been bombed in the United States. Daniel
Andreas San Diego is on the FBI’s Most Wanted list for having allegedly planted bombs against two
biotechnology companies that had commercial ties with Huntington Life Sciences; he remains on the
run.2163 FBI believes San Diego to be “involved with” the Stop Huntington Animal Cruelty group, hence
the inclusion of this case under the domestic terrorists and extremists section.2164 More specifically, he is
wanted for the bombing of the Chiron Life Science Center and the Shaklee Corporation building, both in
California in 2003.2165,2166,2167 The bomb used against Shaklee Corporation was designed to produce
shrapnel through the addition of nails around the explosive charge, while the attack against Chiron
included a secondary bomb in a possible attempt at targeting first responders.2168
2159 Ackerman G (2003) “Beyond Arson? A Threat Assessment of the Earth Liberation Front,” Terrorism and Political Violence
Vol. 15, 4.
2160 Phillips L (2012) “Anarchists attack science,” Nature (News) 485, no. 561, http://www.nature.com/news/anarchists-attack-
science-1.10729 Accessed September 11, 2015.
http://www.nature.com/news/2011/110822/full/476373a.html. Accessed September 11, 2015.
2162 Ibid.
2163 Federal Bureau of Investigation (FBI), “New Most Wanted Terrorist: First Domestic Fugitive Added to List,” April 21,
2009, https://www.fbi.gov/news/stories/2009/april/wanted_042109. Accessed August 27, 2015.
2164 Ibid.
2165 Ibid.
2166 Federal Bureau of Investigation (FBI), “Terrorism 2002-2005,” p.9-10.
2167 Rodriguez M, Chong JR, Krikorian G (2003) “Suspect is Sought in Bombings,” Los Angeles Times,
http://articles.latimes.com/2003/oct/10/local/me-warrant10. Accessed 28, 2015.
2168 Federal Bureau of Investigation (FBI), “New Most Wanted Terrorist: First Domestic Fugitive Added to List.”

In addition, eco-radical groups in Latin America and Europe have targeted nanotechnology researchers by
sending mail bombs to the researchers’ laboratories.2169,2170 These groups have issued propaganda against
synthetic biology research, and as such could conceivably target biological researchers in the future.2171
So far, these groups have not targeted U.S. laboratories or U.S. researchers.
16.5.4.3 Covert entry (physical):
ALF has repeatedly covertly entered laboratories (see Sec. 16.1), although none of the facilities breached
were secured at the current high containment or Biological Select Agents and Toxins levels.
16.5.4.4 Covert entry (cyber):
Cyber-operations by terrorists have so far been largely limited to simplistic attacks, such as website
defacement.2172 Director of National Intelligence James R. Clapper remarked in the latest 2015
unclassified Worldwide Threat Assessment of the U.S. Intelligence Committee that: “terrorist groups will
continue to experiment with hacking, which could serve as the foundation for developing more advanced
capabilities. Terrorist sympathizers will probably conduct low-level cyber attacks on behalf of terrorist
groups and attract attention of the media, which might exaggerate the capabilities and threat posed by
these actors.”2173 Based on these remarks, domestic extremist or terrorist groups are currently judged
incapable of sabotaging laboratories through the hijacking of engineering control systems, which would
require a sophisticated cyber attack. In addition to the complexity of the required attack code needed to
interact with the engineering control systems, such an attack could require the penetration of an air-
gapped network if the engineering controls are on an isolated intranet.
16.5.4.5 Theft of pathogen:
Although ALF has covertly entered laboratories, no open source reports documented a case of theft of
pathogen (see Sec. 16.3).
16.5.4.6 Theft of material, animals, or information:
ALF has stolen research documents from laboratories and individuals in a direct attempt to disrupt
research they oppose.2174 ALF has stolen animals from facilities, but they have not stolen from high
2169 Phillips L (2012) “Anarchists attack science,” Nature (News) 485, no. 561, http://www.nature.com/news/anarchists-attack-
science-1.10729 Accessed September 11, 2015.
http://www.nature.com/news/2011/110822/full/476373a.html. Accessed September 11, 2015.
2171 Ibid.
2172 Theohary C, Rollins J (2011) “Terrorist Use of the Internet: Information Operations in Cyberspace,” Congressional
Research Service, https://www.fas.org/sgp/crs/terror/R41674.pdf. Accessed August 3, 2015.
2173 Director of National Intelligence, James R. Clapper, Statement for the Record- Worldwide Threat Assessment of the U.S.
Intelligence Community, Senate Armed Services Committee, February 26, 2015, p. 3,
http://www.dni.gov/files/documents/Unclassified_2015_ATA_SFR_-_SASC_FINAL.pdf. Accessed August 3, 2015.
2174 “Lab Records, Dogs Stolen From Baby Fae Surgeon,” Los Angeles Times, August 16, 1988, retrieved at Orlando Sentinel,
http://articles.orlandosentinel.com/1988-08-16/news/0060190259_1_baby-fae-linda-university-loma-linda. Accessed August
3, 2015.

contentment facilities (see Sec. 16.1). They have also stolen animal cages to aid in the exfiltration of
animals from facilities (see Sec. 16.1).2175
16.5.4.7 Sabotage
ALF and other animal rights extremist groups have typically sabotaged buildings they have broken into.
Five cases of laboratory sabotage (excluding arson) are documented in all five were carried out by ALF.
These acts were not covert and were not intended to injure. However, since these groups often destroyed
machinery to prevent continued experiments, the risk of an accidental release from sabotage cannot be
ruled out.
Elicitation is likely to have been carried out in planning domestic terrorist and extremist attacks, although
this type of information is rarely documented in open sources. FBI has noted that “ALF activists will not
merely attack a university where animal research is conducted, but rather will attempt to locate the
specific laboratory at the university where the research is being conducted […].”2176 Elicitation is one
potential method that can be used find out which laboratory is conducting animal research, but is not the
sole means of doing so. A memoir by an ALF member makes mention of students providing information
to ALF “moles” during the planning phase of a laboratory attack; the use of the term “mole” suggests that
the students were being elicited by ALF.2177 As such, domestic extremist groups appears to have used
elicitation on more than one occasion.
ALF is believed to have either subverted employees or inserted operatives on several occasions. FBI has
stated that animal rights extremists have “obtain[ed] proprietary or confidential information about
intended victim companies through theft or from sympathetic insiders.”2178 Camera footage from one
laboratory break-in perpetrated by the group showed members with access keys, while another break-in
without signs of forced entry left police wondering if the group had a physical key.2179,2180 A memoir by
an ALF member mentions “moles” providing information to the group.2181 An account by a supporter of
2175 Statement of Senator David Vitter, “opening statement,” Oversight on Eco-terrorism specifically examining the Earth
Liberation Front (“ELF”) and the Animal Liberation Front (“ALF”), U.S. Senate Committee on Environment & Public
Works, May 18, 2005, http://www.epw.senate.gov/pressitem.cfm?party=rep&id=237834,
http://www.epw.senate.gov/hearing_statements.cfm?id=237836. Accessed August 11, 2015.
2176 Federal Bureau of Investigation (FBI), “Terrorism 2000/2001,” p. 27.
2177 Anonymous, Memories of Freedom: Western Wildlife Unit of the Animal Liberation Front, p. 21,
http://theanarchistlibrary.org/library/western-wildlife-unit-of-the-animal-liberation-front-memories-of-freedom.pdf.
Accessed October 15, 2015.
2178 John E. Lewis, Deputy Assistant Director, Federal Bureau of Investigation (FBI), Testimony Before the Senate Judiciary
Committee, Washington, U.S.A, May 18, 2004, https://www.fbi.gov/news/testimony/animal-rights-extremism-and-
ecoterrorism. Accessed August 11, 2015.
2179 McGlynn A (2009) “Activist who refused grand jury testimony now charged with conspiracy,” Lancaster Online,
http://lancasteronline.com/your_news/community/activist-who-refused-grand-jury-testimony-now-charged-with-
conspiracy/article_3e187816-29c4-5c46-89aa-6edfbf8c8cfc.html?mode=jqm. July 13, 2015.
2180 Sorensen E “Activists vandalize WSU labs, release research animals,” The Spokesman-Review, A1, A7, retrieved at: Animal
Liberation Frontline, Animal Liberation Front Press Clippings 1984-1994, p.20-21, http://animalliberationfrontline.com/wp-
content/uploads/2014/10/ALF-News-Article-Collection.pdf. Accessed October 15, 2015.
2181 For example: "ALF moles followed up on leads of other potential targets, and searched veterinary medicine files for possible
future actions." In: Anonymous, Memories of Freedom: Western Wildlife Unit of the Animal Liberation Front, p.15, 17, 19,
21, http://theanarchistlibrary.org/library/western-wildlife-unit-of-the-animal-liberation-front-memories-of-freedom.pdf.

the group talks of “the A.L.F.'s source inside the lab” when discussing another group action.2182 In the
U.K., animal rights extremists have subverted a civil servant at the Driver and Vehicle Licensing Agency
to obtain addresses of the individuals they were targeting.2183 The individual apparently provided such
information out of sympathy for animal rights protests; his legal defense argued that he had “believed the
information would be used for lawful protest.”2184 As such, domestic extremist groups appear to have
subverted employees on several occasions.
In addition to the probable ALF cases noted under the “subversion of employee” entry above, R.I.S.E.’s
co-founder Stephen J. Pera probably joined a research group to gain access to pathogen-growing
equipment (see Sec. 16.5).
…Pathogen point-source release from lab: No such cases were found in open source reporting,
although the potential for an event of this type cannot be excluded.
…Release of infected laboratory animals from the laboratory: ALF has released lab animals on
numerous occasions, and has also exfiltrated animals themselves out of the lab, as “animal
liberation” is one of the group’s top priority (see Sec. 15.1). In a 1987 case, animal rights
extremists calling themselves the Band of Mercy stole 11 cats infected with Toxoplasma gondii
and other uninfected animals from a research laboratory.2185 The members reportedly knew that
the cats were infected at the time of the theft, but the group reportedly gave assurances that the
cats had been put under veterinary care after the break-in.2186 In a subsequent 1989 case, ALF
stole mice that were infected with cryptosporidium from a research laboratory.2187 The
perpetrators claimed in a press release that “absolutely no animals were released into the
community,” that “all animals were carefully transported to safe houses,” and that “the infected
mice were […] being treated.”2188 Based on these historical examples, the possibility that
infected animals could be released from the laboratory cannot be ruled out. The potential
consequences of such a release would be contact of infected animal with people, other lab
animals, and/or wild animals that could lead to an outbreak.
…Cross-contamination of laboratory animals: This event has apparently not occurred previously,
although ALF has mixed animal cages before in an attempt to disrupt experiments (see Sec. 15.1).
Should some of the animals be infected with a contagious disease, the practice could cross-
2182 “Blast from the Past- ‘80s Lab Raids,” No Compromise 15, http://www.nocompromise.org/issues/15blast_past.html.
2183 Fenton B (2004) “DVLA mole jailed for aiding guinea pig farm activists,” The Telegraph,
http://www.telegraph.co.uk/news/uknews/1475082/DVLA-mole-jailed-for-aiding-guinea-pig-farm-activists.html. Accessed
October 15, 2015.
2184 Ibid.
2185 Schneider K (1987) “Theft of Infected Cats From U.S. Lab Spurs Alert,” The New York Times,
http://www.nytimes.com/1987/08/25/us/theft-of-infected-cats-from-us-lab-spurs-alert.html. Accessed October 15, 2015.
2186 As given in an account “intended to represent the views of the United States Animal Liberation Front and its members.”
Ingrid Newkirk, Free the Animals: The Amazing True Story of the Animal Liberation Front (New York: Lantern Books,
2000). p. 339-355, front matter, and pictures.
859692&hl=en. Accessed June 26, 2015.
2188 Animal Liberation Frontline, Animal Liberation Front Press Clippings 1984-1994, p. 29,
http://animalliberationfrontline.com/wp-content/uploads/2014/10/ALF-News-Article-Collection.pdf. Accessed October 15,
2015.

contaminate laboratory animals. Whether ALF would deliberately cross-contaminate facilities,
given that the outcome would likely cause harm to some uninfected animals and, hence, violate
ALF guidelines is unclear. However, some ALF members have in practice treated animals within
the raided facilities as already dead, justifying acts that impede the operation of the facility
despite the fact that these actions also pose disproportionate risk to the held animals. This attitude
is most visible in ALF attacks against mink farms, where the animals released have very little
chance to survive in the wild and often end up dead on roads.2189,2190,2191,2192
…Infection of outside animal (Wild or domestic): No such cases were found in open source
reporting, although the 30 infected mice released in the aforementioned 1989 ALF raid had the
potential to spread cryptosporidium for a week to ten days, both through direct contact and
through mice feces.2193
…Infection of lab worker: No such cases were found in open source reporting. A website
dedicated to the ALF alleges that “vials of infectious serum were removed from a refrigerator
[and left] to spoil” in one alleged 1998 break-in at a private research laboratory.2194 No open
source information is available on this alleged case, and no open source documentation is
available to confirm that an attack against the laboratory took place. Indeed, the incident is not
included in FBI’s public list of domestic terrorist and extremist incidents, unlike other ALF
attacks.2195 The apparent support of such a tactic in pro-ALF circles nevertheless raises the
possibility that pathogen vials could be opened and spread out in a laboratory during an ALF
attack, an event which could lead to the infection of a lab worker or a first responder.
…Infection of public: Only two domestic terrorist or extremist groups (Rajneeshee Cult, R.I.S.E.)
have sought a biological weapons capability. Both the Rajneesh Cult and R.I.S.E. are long-
defunct groups whose history is documented in Se. 15.5. Se. 15.3 and Sec 15.6 provide data on
these groups. Overall, these groups were able to begin a BW program because they could
leverage their access to lab pathogens. Their programs were extremely rudimentary from a
technical standpoint, although the Rajneesh group’s efforts were highly effective. In addition, one
(also long-defunct) right-wing supremacist group placed medical waste near a Jewish
organization as part of a hate crime that threatened infection (see Sec. 15.6).
The number of domestic terrorist or domestic extremist groups or members that are active in the
U.S. is not available in open source reporting, since neither the FBI nor the Department of Justice
release such statistics. However, The New America Foundation maintains a running tally of
“homegrown extremism” incidents since 2001, and as of 2015 they had identified 479
2189 “Thousands of mink freed in B.C. in apparent act of èco-terrorism’,” Vancouver Province, August 27, 2008,
http://www.canada.com/reginaleaderpost/news/story.html?id=7d4845f1-4bc7-4162-bb57-4919aff76869. Accessed August
3, 2015.
2190 Carbery G (2010) “Investigation under way after 5,000 mink freed from farm,” The Irish Times,
http://www.irishtimes.com/news/investigation-under-way-after-5-000-mink-freed-from-farm-1.656730. Accessed August 3,
2015.
2191 Minkfarm drabbad av utsläppta djur,” Småland, October 10, 2010,
http://sverigesradio.se/sida/artikel.aspx?programid=105&artikel=4086537. Accessed August 3, 2015.
2192 “Nuovo blitz degli animalisti Liberati 1.400 visoni da pelliccia,” Gazzetta di Mantova, January 18, 2012,
http://gazzettadimantova.gelocal.it/mantova/cronaca/2012/01/18/news/nuvo-blitz-degli-animalisti-liberati-1-400-visoni-da-
pelliccia-1.3080658. Accessed August 3, 2015.
2193 “Diseased mice freed in arson fires, break-in,” Spartanburg Herald-Journal.
2194 “Laboratory Animal Liberation Campaign,” Animal Liberation Front,
http://www.animalliberationfront.com/ALFront/lab.htm. Accessed August 11, 2015.
2195 For ALF events cited in FBI’s compendiums, see for example the “FBI’s Terrorism in the United States 1999” list.
FBI, “Terrorism in the United States 1998,” p.1-24, https://www.fbi.gov/stats-services/publications/terror_98.pdf. Accessed
August 28, 2015.

“homegrown extremists” involved in 35 carried-out plots of which 26 were lethal incidents, and
131 interdicted plots.2196 Comparing this large number to the few cases of BW-related terrorist
and extremist incidents logged in Sec 15.3, demonstrates that domestic terrorist or extremist
groups rarely seek the capability to infect the public, let alone to carry out acts that have the
potential to cause infection with laboratory-derived pathogens.
No such cases were uncovered in open source reporting. The motive behind a domestic terrorist or
extremist group member deliberately self-infecting would most likely be limited to infecting others (i.e.
not suicide or unsanctioned experimentation). As discussed in the “infection of public” entry above, only
two domestic terrorist or extremist groups have sought to infect others and neither considered self-
infection as a means of doing so.
16.5.5 Assessment of Malicious Act Options for Transnational Terrorists, including State-like
Groups
16.5.5.1 Armed assault
No cases reported involved a U.S.-operated or U.S.-owned lab. One case of armed assault conducted by
transnational terrorists against a non-U.S. lab, a heavily-defended defense-related installation in Yemen,
was found in open source reporting (see Sec. 15.1). In that case, a military hospital was attacked by
members of Al Qaeda in the Arabian Peninsula as part of a broader breach of a military compound, and
the group killed doctors and patients inside.2197 The group later apologized for having done so and
claimed that a fighter had disobeyed orders in targeting the hospital rather than focusing on the military
targets at the compound.2198 Terrorist groups have often launched armed assaults against hospitals
overseas, which often have a diagnostic laboratory.2199 However, this act was typically executed to cause
maximum casualties and/or to take many hostages at once, and in no identified cases were pathogens
smuggled out.2200
16.5.5.2 Bombing or arson
No cases reported involved a U.S.-operated or U.S.-owned lab. Transnational terrorists have carried out
several bombing attacks against non-U.S. labs, including the aforementioned attack against one heavily-
defended defense-related installation in Yemen (which was a combined suicide car bomb – armed assault
attack). Sec. 15.1 contains a summary of three such cases. Numerous additional cases of bombings
targeting hospitals have been described in open sources.2201
2196 The New America Foundation International Security Program, “Homegrown Extremism 2001-2015,”
http://securitydata.newamerica.net/extremists/analysis.html, http://securitydata.newamerica.net/extremists/deadly-
attacks.html, http://securitydata.newamerica.net/extremists/terror-plots.html,
http://securitydata.newamerica.net/extremists/methodology.html. Accessed June 30, 2015.
2197 Nasser Arrabyee, Ben Hubbard, “Attack on Yemen’s Defense Headquarters Is Linked to Al Qaeda,” The New York Times,
December 6, 2013, http://www.nytimes.com/2013/12/07/world/middleeast/yemen-attack.html. Accessed August 21, 2015.
2198 The group claimed to be targeting alleged drone control rooms and American experts at the site. Associated Press, “Al
Qaeda Branch in Yemen Regrets Hospital Attack,” Associated Press through The New York Times, December 22, 2013,
http://www.nytimes.com/2013/12/23/world/middleeast/al-qaeda-branch-in-yemen-apologizes-for-attack-on-hospital-at-
defense-ministry.html. Accessed August 21, 2015.
2199 Boaz Ganor, Miri Halperin Wernli, “Terrorist Attacks against Hospitals Case Studies,” International Institute for Counter-
Terrorism, October 27, 2013, p. 1-32, http://www.ict.org.il/Article/77/Terrorist-Attacks-against-Hospitals-Case-Studies.
July 13, 2015.
2200 Ibid.
2201 Ibid.

16.5.5.3 Covert entry (physical) for theft of pathogens or information:
No such cases were found in open source reporting, although numerous transnational terrorist groups
have carried out covert infiltrations to hit their targets. The lack of cases of covert entry is, therefore,
simply a byproduct of the relative lack of attacks against laboratories. According to a National Research
Council publication, unspecified terrorist websites have “suggested that their operatives can pose as
students to gain access to university laboratories and remove hazardous chemical, biological, or
radiological agents.”2202 Although this statement suggests that at least some low-level interest among
transnational terrorists in covert entries at university laboratories exists, whether U.S. laboratories are
considered for attack is unclear.
16.5.5.4 Covert entry (cyber)
Refer to the cyber covert entry in the Lone Outsider overview for a general overview of the potential
networks attacked and of the malicious acts that could be facilitated through successfully penetrating
these networks.
Cyber-operations by terrorists have so far been largely limited to simplistic attacks, such as website
defacement.2203 Director of National Intelligence James R. Clapper remarked in the latest 2015
unclassified Worldwide Threat Assessment of the U.S. Intelligence Committee that: “terrorist groups will
continue to experiment with hacking, which could serve as the foundation for developing more advanced
capabilities. Terrorist sympathizers will probably conduct low-level cyber attacks on behalf of terrorist
groups and attract attention of the media, which might exaggerate the capabilities and threat posed by
these actors.”2204 Based on these remarks, transnational terrorist groups are currently judged incapable of
sabotaging laboratories through the hijacking of engineering control systems, which would require a
sophisticated cyber attack. In addition to the complexity of the required attack code needed to interact
with the engineering control systems, such an attack could require the penetration of an air-gapped
network if the engineering controls are on an isolated intranet.
16.5.5.5 Sabotage
No known instances of sabotage of a laboratory by transnational terrorists were found in open source
reports (see Sec. 15.3). Should a group launch an armed assault against a laboratory, they are likely to
carry out overt sabotage once within a laboratory, which may in turn lead to a breach in containment. For
instance, when transnational groups capture a high-profile location and/or hold large numbers of hostages,
they often set explosives to complicate hostage rescue.2205,2206,2207 A group that captures a laboratory as
part of a negotiating strategy should be expected to set explosives. The charges could be detonated, either
by the terrorists, or accidentally as part of a rescue attempt gone wrong. This event has occurred before,
2202 National Research Council of the National Academies, Prudent Practices in the Laboratory: Handling and Management of
Chemical Hazards, Updated version (Washington: The National Academies Press, 2011), p. 260.
2203 Theohary C, Rollins J (2001) “Terrorist Use of the Internet: Information Operations in Cyberspace,” Congressional
Research Service, https://www.fas.org/sgp/crs/terror/R41674.pdf. Accessed August 3, 2015.
2204 Director of National Intelligence, James R. Clapper, Statement for the Record- Worldwide Threat Assessment of the U.S.
Intelligence Community, Senate Armed Services Committee, February 26, 2015, p. 3,
http://www.dni.gov/files/documents/Unclassified_2015_ATA_SFR_-_SASC_FINAL.pdf. Accessed August 3, 2015.
2205 A few recent, high profile, examples include: Lamine Chikhi, Bate Felix, “Sahara Islamists take hostages, spreading Mali
war,” Reuters, January 16, 2013, http://www.reuters.com/article/2013/01/16/us-sahara-crisis-idUSBRE90F1JJ20130116.
2206 `When kids bury kids’: Russia remembers 130 victims of Nord-Ost terror act 10 years on,” Russia Today, October 23, 2012,
http://rt.com/news/nord-ost-terror-anniversary-827/. Accessed July 15, 2015.
2207 “Ì don’t feel guilty’: Single surviving Beslan terrorist unrepentant 10 years after tragedy,” Russia Today, September 1,
2014, http://rt.com/news/184044-only-surviving-beslan-terrorist/. Accessed July 15, 2015.

for example when the terrorists detonated charges during the Beslan hostage crisis.2208 An internal
explosion could breach containment walls, damage filtration and air pressure systems, breach animal
pens, and breach infected waste storage and disposal systems. In turn, hostages, responders, the general
public, and the terrorists themselves could be exposed to pathogens.
Elicitation is likely to have been carried out in planning some transnational terrorist attacks, although this
type of information is not documented in open sources.
16.5.5.7 Subversion of employee
Same as for “Covert entry,” mutatis mutandis.
Same as for “Covert entry,” mutatis mutandis.
…Pathogen point-source release from lab: No such cases were found in open source reporting.
However, transnational groups have often planted explosives in captured buildings, including
hospitals, in an effort to complicate hostage rescue.2209 As noted above, a scenario where a group
captures a laboratory to use as a negotiating strategy may degenerate into a pathogen point-source
release even if this was not the end goal of the terrorists. This outcome could be deliberately
occasioned by the terrorists themselves, or the result of a law enforcement response gone awry.
…Release of infected lab animals from/within the lab: No such cases were found in open source
reporting.
…Cross-contamination of laboratory animals: No such cases were found in open source

reporting.
…Infection of lab animals outside of containment: No such cases were found in open source
reporting.
…Infection of outside animal (Wild or domestic): No such cases were found in open source
reporting, although Al Qaeda apparently considered some type of attack against U.S. agriculture
given that “hundreds of pages of U.S. agricultural documents” were apparently recovered from Al
Qaeda hideouts in Afghanistan.2210,2211,2212
2208 Ekaterina Stepanova, “From Dubrovka to Beslan: Who is learning faster?,” PONARS Policy Memo 347, November 2004,
p.3, http://csis.org/files/media/csis/pubs/pm_0347.pdf. Accessed July 15, 2015.
2209 Ibid.
2210 Reported by: Susan Collins, “Opening Statement,” in Agroterrorism: The Threat to America’s Breadbasket, Senate
Committee on Governmental Affairs, S.Hrg. 108-491, Nov. 19, 2003, http://frwebgate.access.gpo.gov/cgi-
bin/getdoc.cgi?dbname=108_senate_hearings&docid=f:91045.wais.pdf. Accessed July 14, 2015.
2211 Jim Monke, “Agroterrorism: Threats and Preparedness,” CRS Report for Congress, March 12, 2007, p.1-56,
https://www.fas.org/sgp/crs/terror/RL32521.pdf. Accessed July 14, 2015.
2212 R. Goodrich Schneider et al., “Agroterrorism in the U.S.: An Overview,” University of Florida Food Science and Human
Nutrition Department, Florida Cooperative Extension Service, FSHN0521, August 2009, https://edis.ifas.ufl.edu/fs126.

…Infection of lab worker: No such cases were found in open source reporting.
…Infection of public: Very few transnational groups have pursued a BW program, but the ones
that did intended to kill Americans (see Sec 16.3 through 16.6). Only three transnational terrorist
groups (Al Qaeda Central, Jemaah Islamiyah, and Aum Shinrikyo), have sought a biological
weapons capability based on open source reporting. Sec 16.4 and Sec 16.6 account for other
lesser activities (such as empty threats of use, or the use of biological waste to spike explosive
shrapnel) and “false positives” (groups that were initially believed to have sought BW, but where
a reassessment of the evidence has disputed the prior accounts). Of these three transnational
groups, only Al Qaeda is currently likely pursuing a biological weapons capability. Jemaah
Islamiyah’s membership, including its core leadership and all known BW-program members, has
been decimated in recent years. Aum Shinrikyo’s WMD program has been dismantled.
Furthermore, no credible open source reports exist that would indicate that new groups, such as
ISIL or the al-Nusra Front, are attempting to obtain pathogens for use as biological weapons (see
Sec. 16.3 and 16.4 on terrorist interest in BW, and Sec. 16.12 on ISIL).
Published studies that attempt to predict which future groups are most likely to pursue BW have
produced widely varying results, indicating that no credible method exists to predict future
terrorist interest in a BW capability. These efforts can be divided into two categories: quantitative
searches for predictive indicators, and more qualitative organizational learning studies. The
aforementioned quantitative study analyzing a dataset of 395 terrorist organizations active in the
1998-2005 period showed that only 23 had reportedly pursued some type of CBRN capability.2213
The authors concluded that the larger the organization, the greater the number of attacks of any
type it had previously launched, and the greater the number of allied groups the organization had,
the more likely the organization was to pursue chemical, biological, radiological or nuclear
(CBRN) weapons.2214 They further noted that the ideologies of groups interested in CBRN
weapons varied widely and reported that in particular, religious ideology was not a significant
predictor for whether or not a group would seek CBRN weapons.2215 Based on these findings, the
authors noted “the apparent ascendancy of organizational variables (alliance connections,
inexperience, and to [a] less certain extent, organizational size) over other factors, such as the
much touted influence of religion.”2216 However, quantitatively determining whether any of these
conclusions hold true when only biological weapons-seeking groups are analyzed is difficult,
given the tiny sample size of groups who have sought such a capability. For example, one profile-
based predictive model quantifying the risk of certain groups launching CBRN attacks failed to
identify Jemaah Islamiyah as interested in BW; the model predicted a less than 2 percent chance
that the group would pursue CBRN-type attacks.2217 One study attempted to combine quantitative
work with qualitative studies in part to move beyond this limitation. This 2014 study conducted
by START researchers Gary Ackerman and Markus Binder used three different methods to
generate a “top ten” list of current biological non-state adversaries.2218 Using a new self-created
dataset, they generated a quantitative model and produced a first ranking.2219 This list was then
compared to results from their own assessment based on the historical cases in their dataset, as
2213 Ibid.
2214 Ibid.
2215 Ibid.
2216 Ibid.
2217 Alexandra Pocek Joosse, H. Brinton Milward, “Organizational Versus Individual Attribution: A Case Study of Jemaah
Islamiyah and the Anthrax Plot,” Studies in Conflict & Terrorism 37 (2014): p. 237, p. 253fn.4.
2218 Gary Ackerman, Markus Binder (for START), “Anatomizing the Behavior of Chemical and Biological Non-State
Adversaries,” PASCC Semi-Annual Workshop on Strategic Stability and WMD, Washington, U.S.A., December 5, 2014, p.
11, http://csis.org/files/attachments/141205_Ackerman_Slides_0.pdf. Accessed July 13, 2015.
2219 Ibid.

well as rankings provided by external subject matter experts.2220 The resultant three lists differed
widely from one another. In conclusion, no rigorous method exists for identifying terrorist groups
who have not yet been caught pursuing BW but who are likely to do so in the future.
Although determining whether the group will pursue a BW program in the future is difficult, ISIL
is of particular concern given its enormous resources, the fact that other groups abroad have
sworn fealty to the organization, its ability to recruit or coerce engineers and scientists both in
Syria and Iraq and in Western countries, its anti-U.S. rhetoric and actions, its predilection for
carrying out atrocities that it knows will generate mass media attention, its apocalyptic beliefs,
and its apparent use of chemical weapons. As of August 2015, ISIL had already carried out or
inspired 55 plots against the West (including Australia), and was linked to 14 plots against the
U.S. in particular.2221 Sec. 16.12 summarizes relevant information on the group.
16.5.5.10 Deliberate self-infection
No such cases were uncovered in open source reporting. The motive for a deliberate self-infection by a
transnational terrorist member would most likely be limited to infecting others. As explained in the
section above, only three transnational terrorist groups have sought to infect others. None are known in
open sources to have planned for self-infection as a means of doing so (see Sec. 16.3 and 16.4 on terrorist
interest in BW, and Sec. 16.12 on ISIL).2222
16.5.6 Assessment of Malicious Acts Options for Foreign Intelligence Entities
16.5.6.1 Armed assault, Bombing or arson, Sabotage
A number of foreign intelligence agencies have the capability to orchestrate an armed assault against a
U.S. lab, or a bombing or an arson attack against a U.S. lab, or the covert or overt sabotage of a U.S. lab.
However, such a direct act would be cause for war and, therefore, be highly unlikely.
16.5.6.2 Covert entry (physical)
No instances of physical covert entry into a U.S. biology lab by an individual or team sent by a foreign
intelligence agency were found in open source reporting, although this event is not outside the realm of
possibility given that foreign intelligence agencies have targeted U.S. labs before to steal research
information. Indeed, this type of incident is unlikely to be captured in open sources, especially as
successful covert entries may go entirely undetected. However, the subversion of an employee or the use
of a cyber-espionage tool is significantly easier to organize from afar, and are, therefore, probably the
preferred options. Reflecting on the differences between a cyber-espionage campaign and the recruitment
and handling of an agent for espionage, the Office of the National Counterintelligence Executive noted
that cyber-espionage was “faster and cheaper” and that it solved the logistical problem of having to
transfer large volumes of documents from an agent to their foreign handler.2223
2220 Ibid.
2221 Chairman Michael McCaul, Committee on Homeland Security, “Chairman McCaul Releases August `Terror Threat
Snapshot’,” August 4, 2015, http://homeland.house.gov/press-release/chairman-mccaul-august-terror-threat-snapshot,
http://homeland.house.gov/sites/homeland.house.gov/files/documents/August%20Terror%20Snapshot_0.pdf. Accessed
August 11, 2015.
2222 Note that terrorist discussion of BW carried out through “martyrdom” (suicide) operations does not imply self-infection as
the chosen vector of spread.
2223 Office of the National Counterintelligence Executive, “Report to Congress on Foreign Economic Collection and Industrial
Espionage, 2009-2011,” p.2.

16.5.6.3 Covert entry (cyber)
A number of cyber-espionage campaigns have been mounted in recent years that have apparently
targeted, inter alia, U.S. research institutions and biopharmaceutical industries. Several such campaigns
were persistent, well-organized, and appeared state-sponsored. Sophisticated cyber-espionage campaigns
that have targeted at least in part the pharmaceutical industry include the “Epic Turla” and “Dragonfly”
cyber-campaigns.2224,2225 The “Epic Turla” campaign saw the infection of “several hundred” computers
across 45 countries, including those of “government institutions, embassies, military, education, research
and pharmaceutical companies.”2226 These incidents raise the possibility that laboratory research could be
stolen by a foreign state or by a criminal group working for a foreign state without the need to carry out a
physical covert entry operation. The “Dragonfly” campaign apparently targeted industrial control devices
controlling pharmaceutical production lines to gain access to sensitive information and potentially steal
“proprietary recipes and production batch sequence steps, as well as […] information that indicate
manufacturing plant volumes and capabilities.”2227 The subversion of pharmaceutical industrial control
systems raises the possibility of plant sabotage and not just espionage, although the “Dragonfly”
campaign did not involve sabotage operations.2228
16.5.6.4 Theft of pathogens
Whether actual pathogen samples, rather than research material, were exfiltrated out of U.S. laboratories
by foreign agents remains unknown.
16.5.6.5 Theft of material or information
Theft of research material has occurred previously at U.S. biology labs. The Soviet Union’s KGB
Directorate T of its First Main Directorate conducted scientific espionage against the West.2229 KGB
agents in the 1980s were tasked to report on select pathogen and toxin research, including influenza, in
addition to other information. KGB agent handlers were especially interested in the “presence and
characteristics of microorganisms with altered properties (new strains resistant to drugs and to the action
of chemical and physical environmental factors, not detectable by standard serodiagnostic methods,
carrying genetic determinants of virulence of heterogeneous microbial species, and capable of
overcoming specific immunity).”2230 Targets for information collection included the National Institutes of
Health, selected due to its research on chemical and biological warfare agent effects.2231 KGB archives
exfiltrated by Mitrokhin reportedly showed the presence of a KGB agent at the conglomerate Du Pont de
Nemours, which conducted work in the chemical, petrochemical, and biomedical sectors.2232 Experts
believe that this collection effort was at least in part conducted in support of the covert Soviet BW
2224 Kaspersky Lab Global Research and Analysis Team, “The Epic Turla Operation: Solving some of the mysteries of
Snake/Uroburos,” SecureList, August 7, 2014, https://securelist.com/analysis/publications/65545/the-epic-turla-operation/.
2225 Kaspersky Lab Global Research and Analysis Team, “Energetic Bear – Crouching Yeti,” July 31, 2014, p. 2,
https://securelist.com/files/2014/07/EB-YetiJuly2014-Public.pdf. Accessed July 14, 2015.
2226 Kaspersky Lab Global Research and Analysis Team, “The Epic Turla Operation: Solving some of the mysteries of
Snake/Uroburos.”
2227 Joel T. Langill, “Defending Against the Dragonfly Cyber Security Attacks,” version 3.0, White Paper, Belden, December
10, 2014, p. 1, 5-8.
2228 Ibid.
Press
2230 Ibid.
2231 Christopher Andrew, Vasili Mitrokhin, The Sword and the Shield: The Mitrokhin Archive and the Secret History of the KGB
(New York: Basic Books, 1999), p. 614 fn. 109.
2232 Ibid

program.2233
The theft of biotechnology and research on pathogens for commercial gain or for unknown purposes by
state actors is believed to be ongoing. One example of an alleged attempted physical theft occurring in a
laboratory outside of the U.S. was reported by Russian Federal Security Service (FSB) officers, who
stated on December 22, 2004 that they had prevented an attempt by foreign intelligence to extract
research data from the Russian biological research institute VECTOR.2234 VECTOR is an institute that
conducts research on extremely dangerous pathogens, and is one of the two official repositories of the
smallpox virus.2235 The regional head of the FSB, Sergei Savchenkov, further remarked that foreign
agents had been specifically targeting microbiology and genetic engineering research.2236 Whether these
claims are true or propaganda is unclear.
Elicitation for espionage purposes has certainly occurred, for instance in support of the aforementioned
KGB information campaigns. Elicitation incidents are however rarely documented in open sources, as a
successful elicitation will not raise the suspicion of the victim and hence is likely to go by unreported.2237
16.5.6.7 Subversion of employee
Employees at U.S. labs have been subverted by foreign intelligence agencies previously, as the
aforementioned KGB cases demonstrate. The KGB office in the U.K. had for their part recruited a lab
assistant in the U.K. under the code name STEP.2238
As in the case of “covert entry (physical),” no such instances were uncovered in open source reporting,
although the possibility of the insertion of an operative into a U.S. lab cannot be ruled out. The subversion
of an employee is by far easier to organize from abroad and, along with cyber-espionage, probably the
preferred options.
Given the rigorous vetting process used by intelligence agencies on their employees, most reckless acts by
a foreign agent or foreign agent team infiltrated into a laboratory can be dismissed. In cases where an
individual is recruited by a foreign intelligence agency in country, the risks of reckless acts may be
similar to an act carried out by a lone insider.
…Infection of outside animals (Wild or domestic): Foreign intelligence agencies might

hypothetically be tasked by an unscrupulous government to covertly infect animals to cause
serious economic damage to a rival state.2239 Despite a number of allegations, no modern cases
Press.
2234 IHS Jane’s, “RIA-Novosti news agency reported on …” Jane’s Intelligence Watch Report – Daily Update, December 23,
2004.
2235 World Health Organization, “World Health Organization inspects Russian smallpox laboratory,” October 25, 2002,
http://www.who.int/mediacentre/news/notes/np7/en/. Accessed August 11, 2015.
2236 IHS Jane’s, “RIA-Novosti news agency reported on …”.
2237 Federal Bureau of Investigation (FBI), “Elicitation Technique.”
2238 Ibid.
2239 Piers Millett, “Antianimal Biological Weapons Program,” Deadly Cultures: Biological Weapons Since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Harvard University Press, 2006), p. 233-235.

have been confirmed in open source reporting. The only confirmed case of anti-animal warfare by
a foreign state is Germany’s covert anti-animal BW operations in World War I, which targeted
pack animals (horses in the United States) in an attempt at disrupting war logistics.2240,2241,2242
…Infection of public: A state willing to violate its international obligations under the 1972
Biological Weapons Convention and international customary norms could hypothetically attempt
to weaponize a pathogen obtained from a laboratory. A number of foreign intelligence agencies,
including the Soviet Union’s and Apartheid South Africa’s, have historically considered using
biological weapons for assassination purposes, i.e. against the public, but in practice have relied
on chemical poisons (including toxins).2243,2244,2245,2246
No useable weapon based on influenza is known to have been developed, and modern programs
run by foreign states (covertly, in contravention to the BWC) are not known to include
weaponization influenza virus strains.2247 However, a U.S. Department of Defense planning
document on responding to pandemic influenza prepared a question-and-answer segment that
summarized the arguments as follows: “many strains of influenza have significant potential for
bioterrorism,” but “because the flu virus mutates so easily many other organisms would be a
better and easier choice for an aggressor to use.”2248 The relatively rapid mutation rate of
influenza is a recognized barrier to hypothetical weaponization.2249
The defunct and pre-BWC Canadian offensive BW program studied influenza A and B in the
1960s.2250 The Canadian BW work was coordinated with the defunct and pre-BWC U.S. and U.K.
2240 France may have reciprocated during WWI by infecting horses destined to Germany, but additional research on the topic is
necessary. W. Seth Carus, “The History of Biological Weapons Use: What We Know and What We Don’t,” Health Security
13, no. 4 (2015): p.233-234.
2241 Mark Wheelis, “Biological sabotage in World War I,” Biological and Toxin Weapons: Research, Development and Use
from the Middle Ages to 1945, Chemical & Biological Warfare Studies No. 18, eds. Erhard Geissler, John Ellis van
Courtland Moon (Oxford University Press, 1999), p. 35-59.
2242 See also the following study, which looked at incidents in North America only: G. A. Ackerman, J. Giroux, “A history of
biological disasters of animal origin in North America,” Scientific and Technical Review of the Office International des
Epizooties (Paris) 25, no. 1 (2006): p. 87.
2243 For example, the Soviet Union considered the use of a Y. pestis dispenser to assassinate Yugoslav leader Josip Tito until
Stalin’s death in 1953 put a halt to the assassination planning. “Stalin’s Plan to Assassinate Tito,” Cold War International
History Project Bulletin 10, March 1998, p. 137. In the post-BWC covert Biopreparat period, the KGB ran a program called
Flute that studied biological weapons suitable for assassinations, notably bioregulatory peptides.
2244 On the Apartheid South African program, see: Chandré Gould, Alastair Hay, “The South African Biological Weapons
Program”; Stephen Burgess, Helen Purkitt, The Rollback of South Africa’s Chemical and Biological Warfare Program,
USAF Counterproliferation Center, April 2001, p. 8-9, 13-16, 21, http://www.au.af.mil/au/awc/awcgate/cpc-
pubs/southafrica.pdf. Accessed September 16, 2015.
2245 On Iraq’s intelligence services and their clandestine labs, see the comprehensive CIA report based on investigations
following the 2003 Iraq War: Central Intelligence Agency (CIA), “DCI Special Advisor Report on Iraq’s WMD, Volume 1:
Iraq’s Intelligence Services,” September 30, 2004, https://www.cia.gov/library/reports/general-reports-
1/iraq_wmd_2004/chap1_annxB.html#sect7. Accessed September 16, 2015.
2246 Central Intelligence Agency (CIA), “DCI Special Advisor Report on Iraq’s WMD, Volume 3: Biological Warfare,”
September 30, 2004, https://www.cia.gov/library/reports/general-reports-1/iraq_wmd_2004/chap6.html#sect4. Accessed
September 16, 2015.
2247 Yannick Pouliot, Jennifer L. O. Sheer, “Influenza,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz,
Raymond A. Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 322-323.
2248 United States Northern Command (USNORTHCOM) Concept of Operations Plan (CONPLAN) 3551-09, Concept Plan to
Synchronize DOD Pandemic Influenza Planning, August 13, 2009, Released under the Freedom of Information Act on June
27, 2013, F-1-3, retrieved at http://www.governmentattic.org/8docs/NORTHCON_CONPLAN_3551-09_2009.pdf.
2249 Ibid.
2250 Donald Avery, “The Canadian Biological Weapons Program and the Tripartite Alliance,” Deadly Cultures: Biological
Weapons since 1945, eds. Mark Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006),
p.405fn.49.

offensive BW programs as part of the Tripartite Alliance.2251 Influenza A was recommended as a
field trial agent because it was “a relatively mild disease in man though it may be temporarily
debilitating […] [it] is dangerous only to the very young and the aged and then only as a result of
secondary bacterial infection.”2252 The defunct and pre-BWC French offensive BW program
considered influenza virus A/PR/8 in 1966 as part of its study on biological incapacitants (the
results and conclusions of this initial study are unknown).2253 These initial studies did not lead to
useable weapons.
The U.S. military appears to have conducted research on human-transmissible influenza for
defensive purposes, particularly for preventing infection of U.S. armed forces with naturally-
occurring influenza.2254 In addition, the U.S. military had concerns that influenza could be used
against U.S. forces. For instance, in June 1961, Colonel Tigertt listed the flu as one out of 40
potential diseases that could be unleashed as part of biological warfare.2255 The U.S. military
conducted volunteer human testing with influenza virus, but influenza was only one of a long list
of potential BW agents assessed.2256 Finally, the U.S. military screened bovine and avian
influenza virus strains for potential as anti-animal warfare agents. Fort Terry (1952-1954)
screened for, researched, and developed certain anti-animal biological warfare agents in its brief
existence (1952-1954).2257 The site held one bovine influenza strain and 34 avian influenza
strains.2258 Avian influenza was selected for its weapons potential, and classed as a second-tier
agent in a three-tier ranking system.2259 However, by the time of the U.S.’ 1969 unilateral
renunciation of biological weapons, no influenza-based weapon was in U.S. stockpiles, as
evidenced by now-declassified documents on materials to be destroyed to comply with the
renunciation announcement.2260 Indeed, the official U.S. Army history of the U.S. BW program
makes no mention of any influenza virus-based weapons, nor of any influenza virus production
lines, nor of any field tests involving influenza virus.2261,2262
Post-1972 offensive BW programs (and therefore post-BWC, covert, programs) apparently did
not include influenza for weaponization. Apartheid South Africa, Iraq, and the Soviet Union ran
offensive BW programs of widely different scale and degree of sophistication in this
2251 Donald H. Avery, Pathogens for War: Biological Weapons, Canadian Life Scientists, and North American Biodefense
(Toronto: University of Toronto Press, 2013), p.113.
2252 Ibid.
2253 Olivier Lepick, “The French Biological Weapons Program,” Deadly Cultures: Biological Weapons since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 125.
2254 Dan Crozier, “History of the Commission on Epidemiological Survey,” Section 1 Part IV, The Armed Forces
Epidemiological Board: The Histories of the Commissions, ed. Theodore E. Woodward (Washington: Borden Institute,
1994), p. 91, 111, 150, 153.. Retrieved at: U.S. Army Medical Department, Office of Medical History, “The Histories of the
Commissions – Contents,” http://history.amedd.army.mil/booksdocs/historiesofcomsn/commission.html. Accessed July 14,
2015.
2255 Ibid, p. 237.
2256 John Ellis van Courtland Moon, “The U.S. Biological Weapons Program,” Deadly Cultures: Biological Weapons since
1945, eds. Mark Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 26.
2257 Piers Millett, “Antianimal Biological Weapons Program,” Deadly Cultures: Biological Weapons Since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Harvard University Press, 2006), p. 226.
2258 Ibid.
2259 Ibid.
2260 See Tab A: Material to be Destroyed (Biological and Toxin), in:
The Secretary of Defense, “Memorandum For the President, National Security Decision Memoranda 35 and 44,” July 6,
1970, Declassified, p.3, http://nsarchive.gwu.edu/NSAEBB/NSAEBB58/RNCBW22.pdf. Accessed June 30, 2015.
2261 U.S. Department of the Army, “U.S. Army Activity in the U.S. Biological Warfare Programs, Volume 1,” p. 50-51.
2262 U.S. Department of the Army, “U.S. Army Activity in the U.S. Biological Warfare Programs, 1942-1977, Volume 2,”
February 24, 1977, Unclassified, p. 102, 124-140, http://nsarchive.gwu.edu/NSAEBB/NSAEBB58/RNCBW_USABWP.pdf.
Accessed June 30, 2015.

timeframe.2263,2264,2265,2266,2267 None are known to have selected influenza as a weapons pathogen,
even though the Soviet Union apparently believed that influenza (“classical fowl plague”) might
be weaponizable, since it remained on a long list of pathogens that KGB agents were supposed to
monitor from Western research.2268,2269,2270,2271
16.6 Attacks against Laboratories
Numerous confirmed cases of attacks against research and medical laboratories have occurred in the U.S.
and abroad, including one operated by a foreign ministry of defense. Eighteen cases between 1990 and
2015 were found in open source literature and documented in Table 16.1. In addition, two cases from
1989 and 1987, which involved the theft of an infected animals, are documented in Table 16.1 because
they are relevant for assessing potential malicious actor motivations and capabilities, and malicious acts.
Laboratory security increased in the 1990s, in part in response to prior incidents. These increases were
highlighted by supporters of the Animal Liberation Front (ALF), a decentralized domestic extremist
group that was responsible for most of the incidents documented below.2272
Press.
2264 United Nations, S/1995/864, October 11, 1995, http://www.un.org/Depts/unscom/sres95-864.htm. Accessed September 27,
2015.
2265 Graham S. Pearson, “The Iraqi Biological Weapons Program,” Deadly Cultures: Biological Weapons since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 177-179, 181.
2266 United Nations Monitoring, Verification and Inspection Commission (UNMOVIC), “Compendium: Chapter V: The
Biological Weapons Programme,” p. 766-1030,
http://www.un.org/Depts/unmovic/new/documents/compendium/Chapter_V.pdf. [Dead link]
2267 Chandré Gould, Alastair Hay, “The South African Biological Weapons Program,” Deadly Cultures: Biological Weapons
since 1945, eds. Mark Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 197-200.
Press.
United Nations Monitoring, Verification and Inspection Commission (UNMOVIC), “Compendium: Chapter V: The
Biological Weapons Programme,” p. 766-1030.
2269 United Nations, S/1995/864.
2270 Graham S. Pearson, “The Iraqi Biological Weapons Program,” Deadly Cultures: Biological Weapons since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 177-179, 181.
2271 Chandré Gould, Alastair Hay, “The South African Biological Weapons Program,” Deadly Cultures: Biological Weapons
since 1945, eds. Mark Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 197-200.
2272 See the following article on a website maintained in support of the Animal Liberation Front: “Laboratory Animal Liberation
Campaign,” Animal Liberation Front, http://www.animalliberationfront.com/ALFront/lab.htm. Accessed August 11, 2015.

Table 16.1. Known attacks against research and medical laboratories, 1989-2015
Date of event Description of event
A suicide car bombing breached a Ministry of Defense complex in Yemen, enabling
2013-12-05 gunmen to enter and attack the Al-Oradi military hospital and laboratory. Al Qaeda in the
Arabian Peninsula (AQAP) claimed responsibility. 2273
Animal rights extremists broke into a laboratory in Milan, Italy. 2274 They released mice and
rabbits, and switched cage labels to sabotage research.2275 According to a worker at the
facility, “The lack of signs of a break-in suggests that the activists may have used an
2013-04-22
illegally acquired electronic card.”2276 Two members chained themselves to the main double
doors. As a result of negotiations with authorities, the group members were allowed to leave
with “fewer than 100 animals.”2277
2012-02-28 A bomb planted against a Pakistani university building damaged, inter alia, a laboratory.2278
2011-08-28 A bomb attack against a Pakistani hospital destroyed a medical laboratory.2279
The Biomedical Research Institute of the University of Hasselt (Belgium) was set
2008-02-01
ablaze.2280,2281
Members of the Animal Liberation Front (ALF) claimed responsibility for vandalizing a
biology laboratory at the Louisiana State University (U.S.A.). The perpetrators exfiltrated 10
2005-04-22 mice, stole some cages, moved mice from cages to cages, and removed mice cage
identifying cards.2282 Overall, members of the ALF had access to a little over 80 mice being
used in experiments.2283
Two laboratories at the University of Iowa (U.S.A.) were vandalized, including through the
pouring of chemicals, and 401 rats and mice were exfiltrated. 2284,2285 Members of the ALF
2004-11-14
claimed responsibility. Video footage released by the group showed the perpetrators had
electronic keys, facilitating access. 2286
2273
Indexed in the START Global Terrorism Database, GTD ID: 201312050016,
http://apps.start.umd.edu/gtd/search/IncidentSummary.aspx?gtdid=201312050016. Accessed June 29, 2015.
2274 Alison Abbott, “Animal-Rights activists wreak havoc in Milan laboratory,” Nature, April 22, 2013,
http://www.nature.com/news/animal-rights-activists-wreak-havoc-in-milan-laboratory-1.12847. Accessed October 15, 2015.
2275 Ibid.
2276 Ibid.
2277 Ibid.
2278 “University campus blown up in Charsadda,” Dawn, March 1, 2012, http://www.dawn.com/news/699336/university-
campus-blown-up-in-charsadda-2. Accessed June 29, 2015. Indexed in the START Global Terrorism Database, GTD ID:
201202280031, http://apps.start.umd.edu/gtd/search/IncidentSummary.aspx?gtdid=201202280031. Accessed June 29, 2015.
2279 Attack mentioned in: “Girls school, shops blown up in Swabi,” Dawn, August 30, 2011,
http://www.dailytimes.com.pk/default.asp?page=2011\08\30\story_30-8-2011_pg7_21 [Dead link]. Indexed in the START
Global Terrorism Database, GTD ID: 201108280002,
2280 Geoff Brumfiel, “Animal-rights activists invade Europe,” Nature (News) 451 (2008): p.1034-1035,
http://www.nature.com/news/2008/080227/full/4511034a.html. Accessed July 17, 2015.
2281 TVL Limburg, archived page at: https://web.archive.org/web/20080208121228/http://www.tvl.be/nl/nieuws/2008-02-
03/brandstichting-alf-op-luc/. Accessed June 29, 2015.
2282 Statement of Senator David Vitter, “opening statement,” Oversight on Eco-terrorism specifically examining the Earth
Liberation Front (“ELF”) and the Animal Liberation Front (“ALF”), U.S. Senate Committee on Environment & Public
Works, May 18, 2005, http://www.epw.senate.gov/pressitem.cfm?party=rep&id=237834,
http://www.epw.senate.gov/hearing_statements.cfm?id=237836. Accessed July 13, 2015.
2283 Ibid.
2284 David Frabotta, “Vandals upend University of Iowa lab,” DVM360 Magazine, January 1, 2005,
http://veterinarynews.dvm360.com/vandals-upend-university-iowa-lab. Accessed June 29, 2015.
2285 Ann McGlynn, “Activist who refused grand jury testimony now charged with conspiracy,” Lancaster Online, November 19,
2009, http://lancasteronline.com/your_news/community/activist-who-refused-grand-jury-testimony-now-charged-with-
conspiracy/article_3e187816-29c4-5c46-89aa-6edfbf8c8cfc.html?mode=jqm. Accessed July 13, 2015.
2286 Ann McGlynn, “Activist who refused grand jury testimony now charged with conspiracy.”

695 mice were exfiltrated out of Wickham Laboratories (U.K.) by two members of the ALF.
2003-12-13 The laboratory held botulinum toxin in the form of Dysport, which is a product used for
therapeutic purposes.2287
Members of the ALF destroyed equipment at the Louisiana State University Inhalation
2003-09-24
Toxicology Research Facility at the School of Veterinarian Medicine (U.S.A.). 2288,2289
Acid and bleach were poured throughout the Sierra Biomedical (U.S.A.) research facility.
2001-11-12
Members of the ALF claimed responsibility. 2290
An incendiary device went off at the White Sands Research Center (U.S.A.), a laboratory
2001-09-20 which used chimpanzees for medical testing. Members of the ALF claimed
responsibility.2291
The Agricultural Hall of the University of Michigan State University was destroyed by
arson (accelerant was found at the scene).2292 Members of the ELF claimed
1999-12-31 responsibility.2293 The attack was launched ostensibly in opposition to a crop genetic
engineering research project, but the Agricultural Hall itself was not a research laboratory
and only contained research data on the project.2294,2295
Animal rights extremists broke into the Avian Health Laboratory of the Washington State
1999-11-19 or 20
University, destroyed equipment, and poured chlorine throughout the facility. 2296
Animal rights extremists into a laboratory in Bellingham run by Western Washington
1999-10-23 or 24 University. Four rabbits and 37 white rats were exfiltrated. 2297 The perpetrators tried to
break into the room where non-human primates were kept, but failed. 2298
Members of the ALF broke into a laboratory owned by private biotechnology company Bio-
1999-08-28 and Devices in Orange County, California and exfiltrated 55 dogs. 2299 Lead X-ray gowns and
29 bottles of medication were stuffed into the facility’s sinks “in an apparent attempt to cause
flooding.”2300
2287 “Veteran animal rights activist jailed after threat in court,” The Guardian, April 30, 2005,
http://www.theguardian.com/uk/2005/apr/30/businessofresearch.animalwelfare. Accessed June 29, 2015.
2288 R. Scott Nolen, “LSU laboratory vandalized; animal extremist group claims responsibility,” JAVMA News, November 1,
2003, https://www.avma.org/News/JAVMANews/Pages/031101a.aspx. Accessed June 29, 2015.
2289 Samantha Sieber, “FBI investigates Vet School break-in,” LSU Reveille, September 25, 2003,
http://www.lsureveille.com/fbi-investigates-vet-school-break-in/article_72954388-993b-5498-9fbf-975a5fde5e4f.html.
2290 Federal Bureau of Investigation (FBI), “Terrorism 2000/2001,” http://www.fbi.gov/stats-
services/publications/terror/terrorism-2000-2001. Accessed June 29, 2015.
2291 Indexed in the START Global Terrorism Database, GTD ID: 200109200006,
2292 Federal Bureau of Investigation (FBI), “Terrorism in the United States 1999,” p. 6, https://www.fbi.gov/stats-
services/publications/terror_99.pdf. Accessed August 28, 2015.
2293 Ibid.
2294 Ibid.
2295 “Four Arrested in 1999 New Year’s Eve Agriculture Hall arson,” MSU Special Report, March 11, 2008, retrieved at
https://web.archive.org/web/20141025083650/http://special.news.msu.edu/ag_hall/index.php. Accessed August 27, 2015.
2296 Mark Rahner, “Equipment is Destroyed at WSU Research Center- Animal Liberation Front Claims Responsibility,” The
Seattle Times, November 22, 1999, http://community.seattletimes.nwsource.com/archive/?date=19991122&slug=2996770.
2297 Janet Burkitt, “Research Animals Taken From Laboratory- Police Suspects Activists Involved in Wwu Break-in,” The
Seattle Times, October 25, 1999, http://community.seattletimes.nwsource.com/archive/?date=19991025&slug=2991001.
2298 Ibid.
2299 Federal Bureau of Investigation (FBI), “Terrorism in the United States 1999,” p. 5.
2300 Ibid.

Members of the ALF broke into twelve laboratories at the University of Minnesota in one
night, causing $2 million in damages and exfiltrating some 100 research animals (pigeons,
rats, and mice).2301 Some of the animals were later found dead in a field, having been
abandoned.2302 One media account provides specific details regarding the break-in at
1999-04-05
researcher Walter Low’s laboratory. 2303 The ALF members poured chemicals on equipment
and papers, and destroyed microscopes, computers, and a particularly valuable tissue
culture.2304 However, they failed to break into the mice storage room in this particular
laboratory.2305
Members of the ALF broke into two animal research buildings on the Michigan State
1992-02-28 University campus (U.S.A.) and set fire to offices, poured sulfuric acid into laboratory
equipment, and opened the cages of minks held for research (the animals did not escape). 2306
Members of the ALF broke into two office buildings and released seven coyotes, 10 mice,
and six minks from two animal research facilities at the Washington State University. 23072308
1991-08-12 or 13
A media report at the time noted that one of the offices broken into had no signs of forced
entry, and that police was wondering whether a key had been used. 2309
Members of the ALF claimed responsibility for the arson of two research buildings,
including a diagnostic laboratory, at the University of Arizona (U.S.A.). They claimed to
have released over 1000 animals from three research facilities. A researcher working at the
1989-04-03 facility noted that 30 missing mice were infected at the time of release with
cryptosporidium.2310 The perpetrators claimed in a press release that “absolutely no animals
were released into the community.” that “all animals were carefully transported to safe
houses,” and that “the infected mice were […] being treated.” 2311
2301 Howard Bell, “Of Mice and Medicine,” Minnesota Medicine, April 2007, http://www.minnesotamedicine.com/Past-
Issues/Past-Issues-2007/April-2007/Feature-April-2007. Accessed August 3, 2015.
2302 Ibid.
2303 Ibid.
2304 Ibid.
2305 Ibid.
2306 “Animal Rights Raiders Destroy Years of Work,” The New York Times, March 8, 1992,
http://www.nytimes.com/1992/03/08/nyregion/campus-life-michigan-state-animal-rights-raiders-destroy-years-of-
work.html. Accessed June 29, 2015.
2307 Eric Sorensen, “Activists vandalize WSU labs, release research animals,” The Spokesman-Review, A1, A7, retrieved at:
Animal Liberation Frontline, Animal Liberation Front Press Clippings 1984-1994, p.20-21,
http://animalliberationfrontline.com/wp-content/uploads/2014/10/ALF-News-Article-Collection.pdf.
2308 Mark Rahner, “Equipment is Destroyed at WSU Research Center- Animal Liberation Front Claims Responsibility.”
2309 Eric Sorensen, “Activists vandalize WSU labs, release research animals,” A7.
2311 Animal Liberation Frontline, Animal Liberation Front Press Clippings 1984-1994, p.29,
http://animalliberationfrontline.com/wp-content/uploads/2014/10/ALF-News-Article-Collection.pdf. Accessed October 15,

2015.

An animal rights extremist group, the Band of Mercy, broke into the Beltsville Agricultural
Research Center in Maryland by cutting through a six-foot link fence and breaking
padlocks.2312 The laboratory was run by the Department of Agriculture and the break-in
resulted in an FBI investigation. 2313 The group members exfiltrated seven African miniature
1987-08-23 pigs, as well as 28 cats.2314 Of these animals, eleven of the cats were infected with
Toxoplasma gondii as part of an experiment.2315 The theft was discovered early morning on
Sunday, August 23, 1987.2316 The members reportedly knew that the cats were infected at
the time of the theft, but the group reportedly gave assurances that the cats had been put
under veterinary care after the break-in.2317
16.7 Biocrimes Committed by Individuals
A “biocrime” is defined here as a criminal act, excluding terrorism, involving a biological substance as
follows: a pathogen, a genetic construct thereof, and medical waste when used with the intent to threaten
infection. Hoaxes and “empty” threats where the actor did not possess or could not be demonstrated to
have possessed a biological substance are not included in this assessment.2318 Incidents involving toxins
such as ricin, which are chemicals, are not included in this assessment. Terrorist incidents, including lone
operator terrorism, are not included in this annex, but are addressed in the terrorism incident annex.
The table below is a compilation of biocrimes, drawing from existing collections in the literature:
• W. Seth Carus’ Bioterrorism and Biocrimes, which documented a total of 16 confirmed

biocrimes (excluding terrorism) where the perpetrator acquired and used a biological agent, and
another 7 that had acquired an agent, from 1900 to 1999,2319
• A short list of incidents compiled by FBI for a 2013 AAAS – AAU – APLU – FBI report on
personnel security programs,2320 and
2312 Keith Schneider, “Theft of Infected Cats From U.S. Lab Spurs Alert,” The New York Times, August 25, 1987,
http://www.nytimes.com/1987/08/25/us/theft-of-infected-cats-from-us-lab-spurs-alert.html. Accessed October 15, 2015.
2313 Ibid.
2314 Ibid.
2315 Ibid.
2316 Ibid.
2317 As given in an account “intended to represent the views of the United States Animal Liberation Front and its members.”
Ingrid Newkirk, Free the Animals: The Amazing True Story of the Animal Liberation Front (New York: Lantern Books,
2000). p. 339-355, front matter, and pictures.
2318 These discarded cases are primarily B. anthracis letter hoaxes and cases where individuals threatened others with what they
claimed were “HIV-infected” sharp objects but where no evidence of actual HIV contamination was found.
2320 American Association for the Advancement of Science (AAAS), Association of American Universities (AAU), Association
of Public and Land-grant Universities (APLU), Federal Bureau of Investigation (FBI), Bridging Science and Security for
Biological Research: Personnel Security Programs, Meeting Report, Washington, United States, August 21-23, 2013, p. 37-
42, http://www.aaas.org/sites/default/files/reports/AAAS-APLU-AAU-
FBI%20report%20on%20personnel%20security%20070114.pdf. Accessed July 13, 2015.

• A synthesis list of CBRN incidents from 1950-2005, listing incidents found in all existing
unclassified incident databases at the time, prepared by Hamid Mohtadi and Antu Murshid.2321
The cases included below are incidents that were reported as “confirmed” in the literature. For all events,
an attempt was made to obtain the primary source material and verify the reported incident directly. When
an incident was reported in open sources but was based solely on a primary source that was not publicly
available, such as the CNS CBRN incident database or interview material reported by the National
Research Council, the incident is reported even when it was not possible to verify the case. These
incidents are clearly flagged as not yet resolved in the text.
The following incidents from 1990 to 2015 have been identified as biocrimes, although this list is unlikely
to be complete. In particular, the coverage of foreign incidents is deficient and is limited to a handful of
high-profile cases. In addition, several unconfirmed and unclear cases have been reported.2322
Table 16.2. Biocrimes, 1990 to 2015

(most recent to
oldest)
Ouyang Xiangyu was a graduate student at Stanford University who allegedly sabotaged lab
Charged 26 mates’ research by killing off their stem cells, and then proceeded to attempt to poison lab
November 2014 mates and herself by putting paraformaldehyde in their water bottles as well as her
own.2323,2324 She pleaded not guilty due to insanity. 2325 The case is ongoing.2326
David Kwiatkowski stole fentanyl syringes to feed his drug habit, and replaced the contents
with a saline solution and his own contaminated blood.2327 He was accused of infecting over
Charged 2012
40 hospital patients with Hepatitis C, including one patient who subsequently died of the
infection, and ended up sentenced to 39 years imprisonment. 2328
2321 The databases in question were: the RAND-St. Andrews Terrorism Chronology, ITERATE, Pinkerton Corporation’s Global
Intelligence Service, the CNS WMD Database, and the MIPT Database. The authors relied on secondary literature
compilations to access data from some of these databases.
Hamid Mohtadi, Antu Murshid, “A Global Chronology of Incidents of Chemical, Biological, Radioactive and Nuclear
Attacks: 1950-2005,” p. 1-5.
Center for Counterproliferation Research, National Defense University, February 2001 Revision),
2323 Melissa Sim, “A*Star scholarship holder Ouyang Xiangyu expelled from Stanford,” April 8, 2015, The Straits Times,
http://www.straitstimes.com/singapore/courts-crime/astar-scholarship-holder-ouyang-xiangyu-expelled-from-stanford.
Accessed September 10, 2015.
2324 “A*Star scholar charged for poisoning labmates’ drinks,” April 2, 2015, TR Emeritus,
https://web.archive.org/web/20150404222631/http://www.tremeritus.com/2015/04/02/astar-scholar-charged-for-poisoning-
labmates-drinks/. Accessed September 10, 2015.
2325 Ibid.
2326 Ibid.
2328 Ibid.

(most recent to
oldest)
Mohsen Hosseinkhani was fired from his job as a cardiology post-doctoral fellow at the
Mount Sinai Medical Center. He then broke into the laboratory on two occasions, stole
July 1 and
equipment as well as “secret scientific material” (including stem cell cultures and
November 28,
antibodies), and sabotaged experiments by switching mice name tags. 2329,2330 He may have
2011
transported stolen materials to Russia. 2331,2332,2333 Once caught, he fled to Iran to avoid
prosecution.2334,2335
A post-doctorate student, Vipul Bhrigu, repeatedly sabotaged the work of another graduate
student working at the same laboratory by spraying ethanol onto his victim’s cell-culture
Charged 2010
media over several months.2336,2337,2338 Once caught, Bhrigu confessed that he “got jealous of
others moving ahead and […] wanted to slow them down.” 2339
Konan Michel Yao stole and attempted to smuggle into the U.S. 22 vials containing DNA
encoding Ebola genes taken from his prior employer, the National Microbiology Laboratory
January 21, 2009
(Canada).2340 He did so in an attempt to transfer his prior research to his new employer. 2341
to
He said he stole the vials on January 21, 2009, during his last day at the lab. 2342 He was then
May 5, 2009
arrested at the U.S.-Canada land border on May 5, 2009.2343 The Canadian lab said they
were reviewing their biosecurity protocol in response to this incident. 2344
2329 Anemona Hartocollis, Al Baker, “Doctor Accused of Crimes Against Mice and Lab,” New York Times – City Room Blog,
December 2, 2011, http://cityroom.blogs.nytimes.com/2011/12/02/doctor-accused-of-crimes-against-mice-and-lab/.
2330 “Lab rat switcher jumps bail, flees to Iran,” Iran Times, http://iran-times.com/lab-rat-switcher-jumps-bail-flees-to-iran/.
2331 Anemona Hartocollis, Al Baker, “Doctor Accused of Crimes Against Mice and Lab.”
2332 Jamie Schram, “Doctor upset over losing hospital fellowship allegedly stole scientific materials, shuffled around lab rats,”
New York Post, December 2, 2011, http://nypost.com/2011/12/02/doctor-upset-over-losing-hospital-fellowship-allegedly-
stole-scientific-materials-shuffled-around-lab-rats/. Accessed June 30, 2015.
2334 Ibid.
2335 “Lab rat switcher jumps bail, flees to Iran.”
2336 Bhrigu denied involvement in earlier cases of potential sabotage; the earliest event of potential sabotage flagged occurred in
December 2009. Bhrigu stated that he had sabotaged his colleague’s work starting in February 2010.
Brendan Maher, “Research integrity: Sabotage!” Nature (News) 467, (2010): p. 516-518,
http://www.nature.com/news/2010/100929/full/467516a.html. Accessed June 30, 2015.
2337 Brendan Maher, “Lab sabotage deemed research misconduct (with exclusive surveillance video),” Nature News Blog, April
27, 2011, http://blogs.nature.com/news/2011/04/lab_sabotage_deemed_research_m_1.html. Accessed June 30, 2015.
2339 Brendan Maher, “Research integrity: Sabotage!” p. 518.
2341 Ibid.
2342 “Winnipeg researcher charged with smuggling Ebola material into U.S.,” CBC News, May 13, 2009,
http://www.cbc.ca/news/canada/winnipeg-researcher-charged-with-smuggling-ebola-material-into-u-s-1.774725. Accessed
June 30, 2015.
2343 Ibid.
2344 Ibid.

(most recent to
oldest)
An attempted theft “targeted at the pathogen collection at the central reference laboratory for
animal health in Indonesia” was “thwarted by security systems installed by the U.S.
May 2007 government.”2345 No further information has been released by the U.S. Department of
State.2346 The motive behind the attempt, as well as whether this incident involved one or
more individuals, is therefore unknown.
A former Cornell University researcher named Qingqiang Yin attempted to smuggle to
China over 250 vials, test tubes, and petri dishes in his luggage. 2347 Some containers were
leaking when they were found.2348 The samples are believed to have contained bacteria and
2002
yeast culture related to enzyme production work that the researcher had been part of. 2349 The
researcher had recently been passed over for re-hire given his reported “poor job
performance,” and Yin was attempting to obtain a job at a laboratory in China. 2350
The James Martin Center for Nonproliferation Studies (CNS) database holds a case
regarding two Kuwaiti prison inmates threatened guards and other inmates with HIV-
contaminated razor blades.2351 CNS classified this as an “incident with possession,” but no
evidence was presented to support this claim. 2352 A UNAIDS Kuwait report stated that in
June 18, 2000
1999, four Kuwaiti prisoners out of 764 tested prisoners were found to be HIV-positive, and
that in 2000, no prisoners out of 1503 prisoners tested were found to be HIV-positive.2353
Based on the existence of HIV-positive prisoners, the account included in CNS’ database
seems possible.
Dr. Larry C. Ford, suspected of having orchestrated the attempted murder of his business
partner, committed suicide.2354 When police searched Ford’s home, they found 266 bottles
and vials of pathogens, including the causative agents of salmonella, cholera, botulism, and
typhoid.2355 Automatic weapons, explosives, and assassination paraphernalia – including the
March 9, 2000
toxin ricin, and a blowgun and dart – were also found.2356 The New York Times article on the
incident presented a number of ultimately vague links between Ford and Apartheid South
Africa’s biological weapons program that had been uncovered; this included allegations that
Ford had smuggled pathogens to South Africa. 2357
2346 Ibid.
2347 Charles Choi, “Lab theft conviction: Former Cornell researcher found guilty of stealing valuable enzymes,” The Scientist,
December 17, 2002, http://www.the-scientist.com/?articles.view/articleNo/21813/title/Lab-theft-conviction/. Accessed
September 7, 2015.
2348 Ibid.
2349 Ibid.
2350 Ibid.
2351 Jason Pate, Gary Ackerman, Kimberly McCloud, “2000 WMD Terrorism Chronology: Incidents Involving Sub-National
Actors and Chemical, Biological, Radiological, or Nuclear Materials,” James Martin Center for Nonproliferation Studies
(CNS), August 13, 2001, http://cns.miis.edu/reports/cbrn2k.htm. Accessed June 30, 2015.
2352 Ibid.
2353 Kuwait, “Epidemiological Fact Sheets on HIV/AIDS and Sexually Transmitted Infections,” UNAIDS, 2004, p.2,
http://data.unaids.org/publications/fact-sheets01/kuwait_en.pdf. Accessed June 30, 2015.
2354 Jo Thomas, “California Doctor’s Suicide Leaves Many Troubling Mysteries Unsolved,” The New York Times, November 3,
2002, p. 1, http://www.nytimes.com/2002/11/03/us/california-doctor-s-suicide-leaves-many-troubling-mysteries-
unsolved.html?pagewanted=1. Accessed June 30, 2015.
2355 Ibid.
2356 Ibid.
2357 Ibid.

(most recent to
oldest)
The CNS database holds a case dated February 25, 2000, where a 17-year old student
allegedly stabbed 37 classmates and a supervisor with a pin allegedly infected with HIV.2358
February 25, 2000
As with the Kuwaiti case listed above, verifying that the case has occurred as described was
not possible.
Three sealed vials reportedly containing samples of coxsackie virus were found on board a
2000
passenger aircraft at the Sydney International Airport (Australia).2359
Two former postdocs from the Harvard Medical School signed an agreement in 2006 stating
December 1999 to that they had stolen research data and materials, including cell lines and genetic material. 2360
January 2000 The theft reportedly occurred over a five day period during the academic winter break,
between 1999 and 2000.2361
Medical waste was deliberately left in several locations in Norwalk and in Stamford
August 1999 (U.S.A.).2362 Two of the six containers sported swastikas and referenced a white supremacist
charged with shooting five people at a Jewish community center. 2363,2364
A burglar stole a physician’s bag containing a vial with a sample of Mycobacterium
tuberculosis.2365,2366 The physician was planning to give the vial to a colleague at a medical
June 28, 1999 conference (according to the media report of this incident, this was not illegal at the
time).2367 Police believed that the burglar did not know the bag contained the vial when the
burglar stole the bag from the physician’s hotel room.2368
2358 Jason Pate, Gary Ackerman, Kimberly McCloud, “2000 WMD Terrorism Chronology: Incidents Involving Sub-National
Actors and Chemical, Biological, Radiological, or Nuclear Materials.”
2359 Hamid Mohtadi, Antu Murshid, “A Global Chronology of Incidents of Chemical, Biological, Radioactive and Nuclear
Attacks: 1950-2005,” July 7, 2006, http://www.ncfpd.umn.edu/Ncfpd/assets/File/pdf/GlobalChron.pdf. Accessed June 30,
2015.
2360 Laurence H. M. Holland, “Couple Admits Cell Line Theft,” The Harvard Crimson, April 17, 2006,
http://www.thecrimson.com/article/2006/4/17/couple-admits-cell-line-theft-in/. Accessed September 10, 2015.
2361 Ibid.
2362 Ibid, p. 37.
2363 “Hate Waste: Task force formed after second container of medical waste found,” The Nevada Daily Mail, August 20, 1999,
2A, retrieved at:
https://news.google.com/newspapers?nid=1908&dat=19990820&id=rDEjAAAAIBAJ&sjid=0NkEAAAAIBAJ&pg=3962,3
2364 Rochelle Rosen, “Medical waste found in lot, Swastika drawn on container at Congregation Beth El school,” The Hour,
August 20, 1999, A1, retrieved at:
https://news.google.com/newspapers?nid=1916&dat=19990820&id=uR9JAAAAIBAJ&sjid=SgYNAAAAIBAJ&pg=3253,
2365 Hamid Mohtadi, Antu Murshid, “A Global Chronology of Incidents of Chemical, Biological, Radioactive and Nuclear
Attacks: 1950-2005,” p. 36.
2366 Associated Press (AP), “San Francisco police seeking TB vial stolen from researcher,” Deseret News, June 29, 1999,
http://www.deseretnews.com/article/704877/San-Francisco-police-seeking-TB-vial-stolen-from-researcher.html?pg=all.
2367 Ibid.
2368 Ibid.

(most recent to
oldest)
Larry Wayne Harris and William Job Leavitt Jr. were arrested by the FBI on charges of
developing and stockpiling a biological agent and conspiring to use it as a weapon. 2369
According to CNN reporting, Leavitt’s defense lawyer argued that the FBI had seized a
substance his client had meant to test and market as an anthrax vaccine. 2370 The substance
seized indeed turned out to be a veterinary vaccine strain of B. anthracis, and Harris was
only charged with violation of his probation in relation to a prior incident.2371 Harris, an
individual associated with several white supremacist groups, had previously been arrested in
Charged 1998
1995 for having forged documents to place an order for Y. pestis from a U.S. laboratory.2372
The order was placed with the American Type Culture Collection in Rockville, Maryland in
the name of a fictitious “Small Animal Microbiology Laboratory,” whose address was in
reality Harris’s home address. 2373 A sales representative who talked with Harris grew
suspicious and called the Centers for Disease Control and Prevention to raise the alarm. 2374
Harris’s home was then raided on May 12, 1995, and he was convicted of fraud on April 22,
1997.2375 Harris’ motivations during each instance is not known.
The New Zealand Ministry of Agriculture and Forestry announced on August 26, 1997 that
rabbit haemorrhagic disease had been detected in New Zealand. 2376 As publicly suspected by
August 26, 1997 the authorities, farmers admitted to having introduced the disease for use as a bio-control
tool; its use had been considered but rejected by the New Zealand Ministry of Agriculture
and Forestry.2377
Laboratory technician Diane Thomspon deliberately infected twelve co-workers at the St.
Paul Medical Center hospital in Dallas, Texas (U.S.A.) with Shigella dysenteriae type 2.2378
She did so by sending out an anonymous email inviting colleagues to eat pastries she had
October 1996 covertly contaminated and left in the break room. 2379 Diane had ready access to a hospital
laboratory holding the Shigella dysenteriae type 2 strain.2380 Diane had previously
contaminated her boyfriend in 1995, also using tainted food, and had then sabotaged his
hospitalization records to prevent a correct diagnosis. 2381
2369 “2 charged with making biological weapons,” CNN, February 19, 1998, http://www.cnn.com/US/9802/19/fbi.arrest.pm/#2.
2370 Ibid.
2371 Lauren Harisson, Jacqueline E. Miller, “Larry Wayne Harris,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds.
Rebecca Katz, Raymond A. Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 383-384.
2373 Lauren Harisson, Jacqueline E. Miller, “Larry Wayne Harris,” p. 383.
2374 Ibid.
2375 Ibid.
2376 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 42;
2377 Ibid, p. 42-43.
2378 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 43-45;
Shellie A. Kolavic, Akiko Kimura, Shauna L. Simons, Laurence Slutsker, Suzanne Barth, Charles E. Haley, “An Outbreak
of Shigella dysenteriae Type 2 Among Laboratory Workers Due to Intentional Food Contamination,” Biological Weapons:
Limiting the Threat, ed. Joshua Lederberg (Cambridge: The MIT Press, 1999), p. 186-192.
2379 Raymond A. Zilinskas, “Diane Thompson: A Case Study,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca
Katz, Raymond A. Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 238.
2380 Ibid.
2381 Ibid.

(most recent to
oldest)
Michael Just attempted to extort British dairies by threatening to contaminate their milk with
Yersinia enterocolitica.2382 He had obtained the pathogen by ordering it from a catalogue
May 1996 supply house.2383 To prove that he was not bluffing, he included test tubes containing
cultures of the pathogen in one blackmail package. 2384 The companies transferred money to
a bank account; Just was arrested attempting to withdraw the funds. 2385
Dr. Richard J. Schmidt, a Louisiana gastroenterologist, deliberately infected a former lover
with HIV and hepatitis using a contaminated hypodermic syringe.2386 He subsequently
August 1994
misled health care professionals by telling them he had tested the victim for HIV with
negative results.2387
Brian T. Stewart was found guilty of having deliberately infected his 11-month old son with
HIV in an attempt to kill the boy.2388 Stewart wanted to avoid paying child support.2389
February 6, 1992 Stewart worked as a phlebotomist, a position which gave him access to the blood of HIV-
positive patients.2390 According to the prosecution, Stewart used this route to obtain
contaminated blood, and then infected his son on February 6, 1992.2391
“Iwan E.” (court identifier for a Dutch man) injected his lover with HIV-contaminated blood
drawn from an HIV-positive friend in June 1992.2392 His lover had just broken up with him;
June 1992
the defense argued this was in a self-defense response to a knife threat made by the
woman.2393
Graham Farlow, inmate at a prison in New South Wales (Australia), infected a prison
July 1990
warder with his HIV-contaminated blood through assault with a syringe. 2394,2395
According to reports by physicians published in early 1991, a French woman attempted
1990?
suicide by injection of HIV-contaminated blood drawn from a friend who had AIDS. 2396
2382 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900.
2383 Ibid.
2384 Ibid.
2385 Ibid.
2386 AAAS, AAU APLU, FBI, Bridging Science and Security for Biological Research: Personnel Security Programs.
2387 Ibid.
2389 Ibid.
2390 Ibid.
2391 Ibid.
2392 Ibid.
2393 Ibid.
2394 Ibid.
2395 Philip D. Jones, “HIV transmission by stabbing despite zidovudine prophylaxis,” Lancet (Letter) 338 October 5, 1991.

(most recent to
oldest)
Kevin T. Birch and James B. Cahoon were convicted of having obtained pathogens under
false pretenses.2397 FBI simply stated that this was for “personal gain;” the media suggested
they had intended to kill a race horse.2398 The two Canadian men pretended to be from “ICM
Science Ltd.” to order freeze-dried Clostridium tetani from the American Type Culture
November 1984
Collection.2399 When ICM Science Ltd. received a copy of the shipping invoice, they
discovered the subterfuge and contacted the police.2400 The two men attempted to effectuate
a second shipment, this time for Clostridium botulinum as well as for Clostridium tetani,
and were arrested.2401
16.8 Terrorist and Extremist Events Tied to Biological Warfare
Table 16.3. Chronology of terrorist and extremist events tied to biological warfare (BW)
Date of event Group Description of event
(most recent to
oldest)
Islamic State of
Foreign Policy journalists report on the obtained contents of one alleged
Iraq and the
August 28, 2014 ISIL member’s laptop.2402 It held over 35,000 files dedicated to Jihad, a
Levant (ISIL,
few of which discussed BW.2403
ISIS, Daesh)
Communist
Party of the
Philippines military claims that NPA used feces to spike explosive devices
Philippines/
2013 to cause sepsis, in what appears to be a modern take on the Viet Cong punji
New People's
stick technique.2404 The NPA denies this.2405
Army
(CPP/NPA)
Revolutionary
Armed Forces A defused FARC gas cylinder bomb reportedly had feces mixed with
May 2012
of Colombia shrapnel in order to cause sepsis upon injury. 2406
(FARC)
2397 Ibid
2398 Ibid.
2399 Ibid.
2400 Ibid.
2401 Ibid.
2402 Harald Doornbos, Jenan Moussa, “Found: The Islamic State’s Terror Laptop of Doom,” Foreign Policy, August 28, 2014,
http://foreignpolicy.com/2014/08/28/found-the-islamic-states-terror-laptop-of-doom/. Accessed June 30, 2015.
2403 Ibid.
2404 “Philippine Army finds human feces, snake venom in wounded soldiers’ wounds,” Nindanao Examiner, September 4, 2013,
http://mindanaoexaminer.com/philippine-army-finds-human-feces-snake-venom-in-wounded-soldiers-wounds/. Accessed
June 30, 2015.
2405 Ibid.
2406 “Army destroys minefield in southwest Colombia,” Colombia Reports, May 17, 2012,
http://colombiareports.com/minefield-and-explosives-found-in-southwest-colombia/. Accessed August 11, 2015.

(most recent to
oldest)
“Indian A 2010 email claiming to be from the “Indian Mujahedeen (Assam)” group
2010 Mujahedeen threatens biological warfare against India unless its demands are met. 2407
(Assam)” However, no evidence exists that this group had or has a BW capability.
Senior AQ member Abu-Salih al Somali authors “Terror Franchise: The
Unstoppable Assassin, TECHS Vital role for its success” sometime after
2009.2408 The document ends with a detailed list of military topics about
After 2009, Al Qaeda (AQ
which the author is requesting the “techs” to research and share instruction
up to 2011 Central)
manuals and videos. BW topics figure prominently on this list, and are
marked as “immediately needed.”2409 The document is captured in the 2011
raid that killed Bin Laden.2410
Al Qaida in the
Islamic
2009 Highly contested news reports of a BW training camp accident.2411,2412,2413
Maghreb
(AQIM)
2407 “Extremists Warn of Biological Strike in India,” Nuclear Threat Initiative Global Security Newswire, October 4, 2010,
http://www.nti.org/gsn/article/extremists-warn-of-biological-strike-in-india/. Accessed June 30, 2015.
2408 David Francis, “Al Qaeda’s Blueprint For How To Start a Homegrown Terror Franchise,” Foreign Policy, May 20, 2015,
http://foreignpolicy.com/2015/05/20/al-qaedas-blueprint-for-how-to-start-a-homegrown-terror-franchise/. Accessed June 30,

2015.
bookshelf?start=1. Retrieved under the “Now Declassified Material” folder: Abu-Salih Al Somali, “Terror Franchise: The
Unstoppable Assassin, TECHS Vital role for its success,” p. 2, 5, 10,
http://www.dni.gov/files/documents/ubl/english/Terror%20Franchise.pdf. Accessed June 30, 2015.
2410 Ibid.
2411 For a critical review of these accounts, see: René Pita, Rohan Gunaratna, Philip Henika, “Al Qaeda in the Islamic Maghreb
(AQIM) and the Alleged Production of the Etiological Agent of Plague,” ASA Newsletter 131 (April 2009): p. 1, 21-22,
http://www.asanltr.com/newsletter/09-2/articles/092a.pdf. Accessed July 17, 2015.
2412 For the accounts themselves, see:
Eli Lake, “Al Qaeda bungles arms experiment,” The Washington Times, January 19, 2009,
http://www.washingtontimes.com/news/2009/jan/19/al-qaeda-bungles-arms-experiment/. Accessed July 14, 2015. And:.
2413 Olivier Guitta, “Al-Qaeda in the Islamic Maghreb: A Threat for the West,” Defence Against Terrorism Review 3, no. 1
(Spring 2010): p. 57-58, http://www.coedat.nato.int/publication/datr/volume5/03-Al-

Qaeda_in_the_Islamic_Maghreb_A_Threat_for_the_West.pdf. Accessed July 14, 2015.

(most recent to
oldest)
The FBI’s complaint filing against Aafia Siddiqui during her trial stated
that at the time of Aafia Siddiqui’s arrest on July 17, 2008, Afghanistan
National Police found “numerous chemical substances in gel and liquid
form that were sealed in glass bottles and glass jars,” as well as “numerous
documents describing the creation of explosives, chemical weapons, and
other weapons involving biological material and radiological agents,”
“documents detailing United States military assets” personal papers
including “descriptions of various landmarks in the United States,
including in New York City.” and “handwritten notes that referred to a
``mass casualty attack’’ and that listed various locations in the United
[Aafia Siddiqui States, including Plum Island, the Empire State Building, the Statue of
July 17, 2008
alleged case] Liberty, Wall Street, and the Brooklyn Bridge.” 2414 The government’s
sentencing submission for the case also holds that her “thumb drive
contained documents […] including: […] discussions of the construction of
chemical and biological weapons.”2415 The prosecution argued that Aafia
Siddiqui’s “conduct was the very definition of a federal crime of
terrorism.”2416 The media reported to the effect that she was a “suspected
al-Qaeda operative;” Saddiqui and her family deny this allegation, and her
trial did not involve an assessment of this accusation. 2417,2418,2419,2420,2421
Since then, the Taliban, the Tehrik-i-Taliban Pakistan, Al Qaeda, and most
recently ISIL have offered (some on multiple occasions) to trade Saddiqui
against hostages.2422,2423,2424,2425,2426,2427
2414 Plum Island is the site of the Plum Island Animal Disease Center, although the prosecution did not elaborate on the alleged
targets.
United States District Court Southern District of New York, United States of America v. Aafia Siddiqui (defendant), “Sealed
Complaint: Violations of 18 U.S.C. §§ 111, 1114, p. 1-3, http://www.justice.gov/archive/opa/pr/2008/August/siddiqui-aafia-
complaint.pdf. Accessed June 30, 2015.
2415 United States District Court Southern District of New York, United States of America v. Aafia Siddiqui (defendant),
“Government’s Sentencing Submission,” Attorney for the United States of America: Preet Bharara, United States Attorney
for the Southern District of New York, Assistant United States Attorneys – of Counsel, Christopher L. LaVigne, David M.
Rody, Jenna M. Dabbs, Case 1:08-cr-00826-RMB, Document 250, Filed August 29, 2010.
http://web.archive.org/web/20120314163620/http://www.nefafoundation.org/miscellaneous/US_v_Siddiqui_usgsentmemo.p
df . Accessed June 30, 2015.
2416 Ibid.
2417 “Dr. Aafia to boycott trial,” The Nation, November 21, 2009, http://nation.com.pk/Politics/21-Nov-2009/Dr-Aafia-to-
boycott-trial. Accessed June 30, 2015.

2418 Benjamin Weiser, “Indictment Hints of Plan to Attack Landmarks,” The New York Times, September 2, 2008,
http://www.nytimes.com/2008/09/03/nyregion/03indict.html?_r=1&.
2419 Petra Bartosiewicz, “Al-Qaeda Woman? Putting Aafia Siddiqui on Trial,” Time, January 18, 2010,
http://content.time.com/time/nation/article/0,8599,1954598,00.html.
2420 Juliane von Mittelstaedt, “America’s Most Wanted: `The Most Dangerous Woman in the World,’ Spiegel Online, November
27, 2008, http://www.spiegel.de/international/world/america-s-most-wanted-the-most-dangerous-woman-in-the-world-a-

593195-druck.html.
2421 “Federal jury convicts Pakistani woman of attempted murder of US personnel,” Jurist, February 4, 2010,
http://jurist.org/paperchase/2010/02/federal-jury-convicts-pakistani-woman.php.
2422 Mushtaq Yusufzai, “Taliban to execute US soldier if Aafia not released,” The News, February 5, 2010,
http://www.webcitation.org/query?url=http%3A%2F%2Fwww.thenews.com.pk%2Ftop_story_detail.asp%3FId%3D27072.
2423 Bill Roggio, “Zawahiri claims al Qaeda is holding US citizen hostage,” Long War Journal – Threat Matrix, December 1,
2011, https://web.archive.org/web/20150103043251/http://www.longwarjournal.org/threat-
matrix/archives/2011/12/zawahiri_claims_al_qaeda_holdi.php.
2424 “Taliban confirm they have Swiss hostages,” Agence France Presse, July 29, 2011, retrieved at The Express Tribune:
http://tribune.com.pk/story/220022/tehrik-i-taliban-say-they-have-swiss-hostages/.

(most recent to
oldest)
The Chinese government alleges that Emeti Yakuf, an alleged terrorist
East Turkistan connected to the East Turkistan Islamic Movement, threatened to use
2008 Islamic biological and chemical weapons to disrupt the 2008 Olympics held in
Movement China, and that he trained group members on making poisons. 2428 This
individual was reportedly killed in a 2012 U.S. drone strike in Pakistan. 2429
Al-Aqsa The group issues a statement claiming that they possess chemical and
June 27, 2006 Martyrs biological weapons, in an attempt to deter Israeli military action. 2430 This
Brigade claim is regarded as spurious.2431
MSNBC reporters state that their initial field tests for botulinum and ricin
toxins came up positive at a site in Iraq used by the group, but that no B.
Ansar al-Islam anthracis was detected; then-Secretary of State Colin Powell had
April 4, 2003
(AAI) previously said the camp held a poison laboratory. 2432 However, in
retrospect, the site does not appear to have produced toxins. The site is not
mentioned in the report of the Iraq Survey Group. 2433
Jemaah Arrest of Riduan Isamuddin, the director of operations for Jemaah
August 2003
Islamiyah Islamiyah who organized for Yazid Sufaat’s transfer to AQ. 2434,2435
Revolutionary
Armed Forces A defused FARC gas cylinder bomb reportedly had feces mixed with
June 2002
of Colombia shrapnel in order to cause sepsis upon injury. 2436
(FARC)
2425 Nima Elbagir, Ingrid Formanek, “Malian troops take key town; humanitarian crisis grows,” CNN, January 21, 2013,
http://www.cnn.com/2013/01/21/world/africa/mali-unrest/.
2426 Jan Lopatka, ed. Alison Williams, “Video of kidnapped Czechs demands release of jailed Pakistani,” Reuters, June 26,
2013, http://www.reuters.com/article/2013/06/26/us-pakistan-czech-kidnapping-idUSBRE95P0XJ20130626.
2427 James Fielding, Marco Giannangeli, “British Aid Worker Executed By Taliban,” Daily Express, October 10, 2013,
http://web.archive.org/web/20101015002351/http://www.dailyexpress.co.uk/posts/view/204533/British-aid-worker-
executed-by-Taliban.
2428 “Èastern Turkistan’ terrorists identified,” China Daily, October 21, 2008, http://www.chinadaily.com.cn/china/2008-
10/21/content_7126503.htm.
2429 Declan Walsh, Eric Schmitt, “Militant Leader Believed Dead in Pakistan Drone Strike,” The New York Times, August 24,
2012, http://www.nytimes.com/2012/08/25/world/asia/us-drone-strikes-kill-18-in-pakistan.html?_r=1.
2430 “Al-Aqsa Martyrs Brigade in Palestine Claims to Have Developed Chemical and Biological Weapons and Threatens Their
Use in Israel,” SITE Monitoring Service Enterprise, June 27, 2006, https://ent.siteintelgroup.com/Jihadist-News/6-27-06-al-
aqsa-martyrs-in-palestine-creates-wmd.html.
2431 Michael Moodie, Markus Binder, “Jihadists and Chemical Weapons,” Jihadists and Weapons of Mass Destruction, eds.
Gary Ackerman, Jeremy Tamsett (Boca Raton: CRC Press, 2009), p. 143.
2432 Preston Mendenhall, “Positive test for terror toxins in Iraq,” MSNBC.com, April 4, 2003,
http://www.nbcnews.com/id/3070394/ns/world_news/t/positive-test-terror-toxins-iraq/#.VXdWckbrJ-A.
2433 Milton Leitenberg, Assessing the Biological Weapons and Bioterrorism Threat, Strategic Studies Institute monograph,
December 2005, p. 26-27, http://www.strategicstudiesinstitute.army.mil/pdffiles/pub639.pdf.

2434 Joel Roberts, “Thailand PM: Hambali Was Plotting,” CBS News, August 17, 2003, http://www.cbsnews.com/news/thailand-
pm-hambali-was-plotting/.
2435 Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?” Paper, Harvard Kennedy School
Belfer Center for Science and International Affairs, January 2010, p. 28,
http://belfercenter.ksg.harvard.edu/publication/19852/al_qaeda_weapons_of_mass_destruction_threat.html.
2436 Mariano C. Bartolome, Maria Jose Espona, “Chemical and Biological Terrorism in Latin America: The Revolutionary
Armed Forces of Colombia,” The ASA Newsletter 03-5, no. 98 (October 31, 2003), http://www.asanltr.com/newsletter/03-
5/articles/035c.htm.

(most recent to
oldest)
Al Qaeda (AQ
Rauf Ahmed is detained in Pakistan, and Yazid Sufaat is arrested in
Central);
December 2001 Malaysia.2437,2438 Pakistan subsequently cuts off FBI access to Rauf Ahmed
Jemaah
in 2003; the latter is now free.2439
Islamiyah
Yazid Sufaat flees Afghanistan for Bogor, Indonesia, to escape from the
Jemaah October 2001 U.S. intervention.2440 He reportedly seeks to set up a new
2001
Islamiyah BW program in-country upon arrival, but fails to recruit a microbiologist at
an Indonesian institute.2441,2442,2443
“At least five envelopes containing significant quantities of Bacillus
September and [Amerithrax anthracis” were mailed to U.S. targets.2444 The attacks killed five and
October 2001 case] sickened seventeen other individuals.2445 FBI concluded that Bruce E.
Ivins, a researcher at USAMRIID (U.S.A.) had sent the letters. 2446
Al Qaeda (AQ Zawahiri launches a BW program in 1999, and hires Rauf Ahmed. 2447,2448
Central); Ahmed establishes a covert laboratory in Afghanistan. 2449 By 2000,
1999-2001
Jemaah Zawahiri recruits Yazid Sufaat.2450 U.S. outing of the Taliban disrupts the
Islamiyah plan and the laboratory is discovered.2451,2452
A Palestinian group (unknown) was reportedly caught in a counterfeiting
1998 to May “Palestinian scheme whereby expired eggs contaminated with salmonella were stamped
2000 Group” with counterfeit stamps and sold.2453 Israeli news reporting on their capture
in May 2000 implied that this was deliberately done to sicken Israelis.2454
2437 Joby Warrick, “Suspect and A Setback in Al-Qaeda Anthrax Case,” The Washington Post, October 31, 2006,
http://www.washingtonpost.com/wp-dyn/content/article/2006/10/30/AR2006103001250.html.
2438 Maria Ressa, “Reports: Al Qaeda operative sought anthrax,” CNN, October 10, 2003,
http://edition.cnn.com/2003/WORLD/asiapcf/southeast/10/10/alqaeda.anthrax/.
2439 Joby Warrick, “Suspect and a Setback in Al-Qaeda Anthrax Case.”
2440 Judith Miller, “U.S. Has New Concerns About Anthrax Readiness,” The New York Times, December 28, 2003,
http://www.nytimes.com/2003/12/28/us/us-has-new-concerns-about-anthrax-readiness.html.
2441 Ibid.
2442 Maria Ressa, “Reports: Al Qaeda operative sought anthrax;”
2443 René Pita, Rohan Gunaratna, “Revisiting Al-Qaìda’s Anthrax Program,”
2444 The United States Department of Justice, “Amerithrax Investigative Summary, Released Pursuant to the Freedom of
Information Act,” February 19, 2010, p. 1, http://www.justice.gov/archive/amerithrax/docs/amx-investigative-summary.pdf.
2445 Ibid.
2446 Ibid.
2447 Alan Cullison, “Inside Al-Qaeda’s Hard Drive,” The Atlantic, September 2004,
http://www.theatlantic.com/magazine/archive/2004/09/inside-al-qaeda-s-hard-drive/303428/.
2448 Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?”
2449 Ibid.
2450 René Pita, Rohan Gunaratna, “Revisiting Al-Qaìda’s Anthrax Program,” CTC Sentinel Vol. 2 Issue 5, May 2009,
https://www.ctc.usma.edu/posts/revisiting-al-qaida%E2%80%99s-anthrax-program.
2451 Ibid.
2452 Rolf Mowatt-Larssen, “How to Get Terrorists to Talk,” The National Interest, February 18, 2015, p.2,
http://nationalinterest.org/feature/how-get-terrorists-talk-12270?page=2.
2453 Jason Pate, Gavin Cameron, “Covert Biological Weapons Attacks against Agricultural Targets: Assessing the Impact against
U.S. Agriculture,” BCSIA Discussion Paper 2001-9, ESDP Discussion Paper ESDP-2001-05, John F. Kennedy School of
Government, Harvard University, August 2001, p.8,
http://belfercenter.ksg.harvard.edu/files/covert_biological_weapons_attacks_against_agricultural_targets.pdf.
2454 Ibid.

(most recent to
oldest)
One Russian newspaper claimed that Salman Raduyev, a prominent
Chechen group
Chechen leader, had threatened to steal biological weapons from ex-Soviet
February 1999 under Salman
biological warfare laboratories unless the government released two
Raduyev
captured women.2455 This report could not be verified. 2456
Two members of the group sent emails threatening to use biological agents
“Republic of
June 1998 against federal officials; no biological agents were uncovered at the time of
Texas”
their arrest.2457
Palestinian A Jordanian newspaper cites a leading figure in the organization as having
April 1998
Islamic Jihad discussed the possibility of using BW.2458 This remains unconfirmed.
National
The ELN detonate an explosive device reportedly spiked with fecal matter
March 6, 1998 Liberation
to cause sepsis upon injury.2459
Army (ELN)
Agar and B. cereus in a petri dish apparently labelled “anthracs” [SIC] and
Counter
“Yersinia” was sent to a Jewish organization in Washington.2460 Whether
Holocaust
1997 this was an anthrax hoax or the group thought the package contained B.
Lobbyists of
anthracis is not known; the package contained a hate latter that further
Hillel
misidentified the petri dish as containing a “chemical warfare” agent. 2461
“Justice
A group calling itself the “Justice Department” mails razors to fur retailers
Department”
1996 in Canada in 1996 which they claim are covered with HIV-infected blood;
[animal rights
whether they really did so is not known. 2462
radical group]
The group ineffectually attempts to disperse botulinum toxin from three
March 15, 1995 Aum Shinrikyo
sprayer-suitcases in the Kasumigaseki metro station (Japan).2463
November 4, The group fails in an assassination attempt involving botulinum toxin
Aum Shinrikyo
1994 mixed with juice.2464
2456 Ibid.
For instance, the Russian think-tank PIR Center does not include this incident in their list of North Caucasus CBRN threat
events. PIR Center, “WMD Terrorism Originated in North Caucasus: Again on the Agenda?” PIR Center Report, April 26,
2013, http://www.pircenter.org/en/articles/1312-wmd-terrorism-originated-in-north-caucasus-again-on-the-agenda.
2457 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 108-109, p. 186.
2458 Ibid.
Armed Forces of Colombia.”
2460 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 110-111;
The B’nai B’rith International Jewish Monthly, Volume 111, (1996), p. 67, https://books.google.com/books?id=V--
2461 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 111.
2462 Ibid.
2463 Richard Danzig, Marc Sageman, Terrance Leighton, Lloyd Hough, Hidemi Yuki, Rui Kotani, Zachary M. Hosford, “Aum
Shinrikyo: Insights into how terrorists develop biological and chemical weapons, second edition,” Center for a New
American Security, December 2012, p. 21,
2464 Ibid.

(most recent to
oldest)
Animal
Liberation A spokesman for the Animal Liberation Front (ALF) claims that bombs
1993 Front (ALF) planted in the U.K. by members of the collective had been purposefully
[animal rights tainted with HIV, but authorities dismiss this account.2465
radical group]
November 18, The group disperses 20 liters of botulinum toxin slurry from a car sprayer
Aum Shinrikyo
1993 in a failed assassination attempt.2466
Following failed attacks with the liquid product, the group sets up a (crude)
1993 Aum Shinrikyo
dry production line for B. anthracis.2467
The group produces some 10 to 20 tons of slurry containing B. anthracis
July-August
Aum Shinrikyo (perhaps not pathogenic), which are then ineffectually released from spray
1993
trucks in some 10 to 20 attacks.2468
The group produces roughly 20 tons of slurry containing B. anthracis
May-June 1993 Aum Shinrikyo (perhaps not pathogenic), and ineffectually sprays the product from the
roof of one of its facilities. 2469
1992 Aum Shinrikyo The group sets up a (crude) liquid production line for B. anthracis.2470
The group produces several hundred tons of slurry as part of their
botulinum toxin production program. 2471 They disseminate this material in
March-July 1990 Aum Shinrikyo
20 to 40 different attempted attacks in this time period, all without
success.2472
Seiichi Endo, the leader of the group’s BW program, harvests C. botulinum
Spring 1990 Aum Shinrikyo
from soil in Japan.2473
S. typhimurium is used to contaminate at least 10 restaurant salad bars in
September 1984 Rajneeshees The Dalles, Oregon (U.S.A.), causing at least 751 people to fall
ill.2474,2475,2476
2466 Richard Danzig et al., “Aum Shinrikyo: Insights into how terrorists develop biological and chemical weapons, second
edition,”
2467 Ibid.
2468 Ibid.
2469 Ibid.
2470 Ibid.
edition,” Center for a New American Security, December 2012, p. 20,
http://www.cnas.org/files/documents/publications/CNAS_AumShinrikyo_SecondEdition_English.pdf>.
2472 Ibid.
edition,” Center for a New American Security, December 2012, p. 18-20,
2474 Thomas J. et. al (1997), “A Large Community Outbreak of Salmonellosis Caused by Intentional Contamination of
Restaurant Salad Bars,” Journal of the American Medical Association 278, no. 5,
http://www.cdc.gov/phlp/docs/forensic_epidemiology/Additional%20Materials/Articles/Torok%20et%20al.pdf;
2475 W. Seth Carus, “The Rajneeshees (1984),” Toxic Terror: Assessing Terrorist Use of Chemical and Biological Weapons, ed.
Jonathan Tucker (Cambridge: The MIT Press, 2001), p. 115;
2476 W. Seth Carus, “Rajneeshees,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A.
Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 383-384.

(most recent to
oldest)
Two Wasco County commissioners were given water deliberately tainted
August 29, 1984 Rajneeshees
with S. typhimurium by Rajneeshees; both fell ill.2477
Reports, based on admissions made by Rajneesh members, of other cult
BW attacks prior to August 1984.2478 These are unconfirmed because none
Early 1984 Rajneeshees of the attacks were successful and because there may have been a desire to
exaggerate wrongdoings by one of the chief organizers (Puja), who was
hated.2479
In an apparent follow-on to the October 10, 1981 incident described below,
Dark Harvest
British police received an anonymous tip that led them to a metal box
October 14, 1981 [eco-radical
allegedly containing B. anthracis.2480 However, unlike in the October 10
group]
incident, the soil did not contain B. anthracis.2481
The eco-radical group “Dark Harvest” took B. anthracis-contaminated soil
from Gruinard Island (a then-contaminated British military WWII site
Dark Harvest
used to test B. anthracis bombs) and spread it on the grounds of Porton
October 10, 1981 [eco-radical
Down in 1981 (Britain’s main biodefense and chemical warfare defense
group]
establishment, and previously the center orchestrating Britain’s biological
weapons program).2482 The soil did contain B. anthracis.2483
Tamil A single unconfirmed account of Tamil “militants” threatening biological
1980s
“militants” warfare.2484
The German-based, now-defunct, Red Army Faction (RAF) reportedly
maintained a botulinum toxin laboratory in Paris, France until it was
uncovered in October 1980.2485 A recent review of this case has cast doubt
Red Army
October 1980 on parts of the underlying story, however, and German authorities
Faction
apparently remain convinced that “no evidence whatsoever [exists] that
members of the `RAF’ had planned or prepared an attack using biological
agents.”2486,2487
2477 W. Seth Carus, “Rajneeshees,” p. 534.

2478 Ibid, p. 534-535.
2480 Ibid.
2481 Ibid.
2483 Ibid.
2484 Ibid.
2485 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 156-157.
2486 Ibid.
2487 The review in question is:
Terence Taylor, Tim Trevan, “The Red Army Faction (1980),” Toxic Terror: Assessing Terrorist Use of Chemical and
Biological Weapons, ed. Jonathan B. Tucker (Cambridge: MIT Press, 2000),

(most recent to
oldest)
POLISARIO; One unconfirmed report of a February 1975 offer by a group called
Basque POLISARIO to coordinate poisoning of water supplies.2488 Even if
February 1975
Fatherland and POLISARIO did make such a threatening offer, no evidence exists that
Liberty (ETA) POLISARIO sought a BW capability.2489,2490
Arrest of two R.I.S.E. founders for having reportedly planned to
January 18, 1972 R.I.S.E. contaminate Chicago’s municipal water system with Salmonella typhi
(causative agent of typhoid fever).2491
16.9 Designated Foreign Terrorist Organizations and Biological Weapons
Table 16.4. Currently designated Foreign Terrorist Organizations and BW

Designated Foreign Publicly Historically BW efforts Notes
Terrorist threatened had a BW believed
Organizations2492 BW? program?2493 ongoing?
Abu Nidal Organization
NO NO
(ANO)
A single news report that police
captured publicly-available reading
material on biological and chemical
Abu Sayyaf Group
NO NO warfare during the capture of six
(ASG)
suspected group members.2494
Whether the group members were
members of ASG or JI is unclear.
Attempted production of BW
agent; launched failed BW attacks.
Aum Shinrikyo (AUM) NO YES NO See the detailed entry below.
Leadership and BW-program
members captured.
One unconfirmed report of a
Basque Fatherland and February 1975 offer by a group
NO NO
Liberty (ETA) called Polisario to coordinate
poisoning of water supplies.2495
2489 POLISARO stands for “Frente Popular para la Liberacion de Saquia el-Hamra y Rio de Oro,” and is a group that seeks to
overthrow Moroccan control of Western Sahara and create an independent state for Sahrawi tribes based on Islamic culture.
2490 Gail H. Nelson, “POLISARIO,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A.
2491 W. Seth Carus, “RISE: A Case Study,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A.
Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 542.
2492 U.S. Department of State, “Foreign Terrorist Organizations,” <http://www.state.gov/j/ct/rls/other/des/123085.htm>.
2493 Where a “BW program” is defined as a military program for the production of a biological pathogen.
2494 Christian Enemark, Disease and Security: Natural plagues and biological weapons in East Asia (Abingdon: Routledge,
2007), p. 106.

Gama’a al-Islamiyya
NO NO
(Islamic Group) (IG)
Reported interest in chemical
Hamas NO NO
poisons.2496
Harakat ul-Mujahidin
NO NO
(HUM)
Hizballah NO NO
Kahane Chai (Kach) NO NO
A single unconfirmed Turkish
Kurdistan Workers Party
NO NO newspaper report of Cobra poison
(PKK) (Kongra-Gel)
smuggling for profit.2497
A single unconfirmed account of
Tamil “militants” threatening
Liberation Tigers of biological warfare in the 1980s.2498
Unconfirmed NO NO
Tamil Eelam (LTTE) Report of LTTE use of chlorine for
chemical warfare.2499 Group has
been defeated.
Unknown
NO, albeit Reports that ELN used feces to
if
reported war spike explosive devices to cause
National Liberation continuing
NO use of sepsis, in what appears to be a
Army (ELN) war use of
biological modern take on the Viet Cong punji
biological
material stick technique.2500
material
Palestine Liberation
NO NO
Front (PLF)
A single unconfirmed April 1998
Jordanian newspaper report citing a
Palestinian Islamic Jihad
NO NO leading figure in the organization
(PIJ)
as having discussed the possibility
of using BW.2501
Popular Front for the
Liberation of Palestine NO NO
(PFLF)
PFLP-General
NO NO
Command (PFLP-GC)
2496 Ibid.
2497 Ibid.
2498 Ibid.
2499 James J. F. Forest, Sammy Salama, “Jihadist Tactics and Targeting,” Jihadists and Weapons of Mass Destruction, eds. Gary
Ackerman, Jeremy Tamsett (Boca Raton: CRC Press, 2009), p. 80.
5/articles/035c.htm.
2501 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 109, 186.

NO, albeit Reports that FARC used feces to
Continued
Revolutionary Armed reported war spike explosive devices to cause
war use of
Forces of Colombia NO use of sepsis, in what appears to be a
biological
(FARC) biological modern take on the Viet Cong punji
material
material stick technique.2502,2503,2504
Revolutionary
Organization 17 NO NO
November (17N)
Revolutionary People’s
Liberation Party/Front NO NO
(DHKP/C)
Shining Path (SL) NO NO
Attempted production of BW
agent, with unknown results. See
al-Qa’ida (AQ) YES YES YES
detailed entry below. Efforts
believed to be ongoing.
By proxy with AQ (central); highly
YES
al-Qaida in the Islamic contested news reports of a BW
(by proxy YES YES
Maghreb (AQIM) training camp accident in
with AQ)
2009.2505,2506,2507
Possibly by proxy with AQ
(central). No information formally
ties this group with AQ’s BW
program. The group reportedly
YES
al-Qa'ida in the Arabian considered contaminating U.S.
(by proxy Unknown
Peninsula (AQAP) food with ricin and cyanide,
with AQ)
although no open source
indications suggest the group
selected this tactic for
operationalization.2508
Armed Forces of Colombia”.
2503 Pablo Esteban Parra Gallego, “IEDs: A Major Threat for a Struggling Society,” The Journal of ERW and Mine Action 13,
no. 3 (Winter 2009), http://www.jmu.edu/cisr/journal/13.3/specialreport/gallego/gallego.shtml
2504 “Army destroys minefield in southwest Colombia,” Colombia Reports, May 17, 2012,
http://colombiareports.com/minefield-and-explosives-found-in-southwest-colombia/.
2505 For a critical review of these accounts, see: René Pita, Rohan Gunaratna, Philip Henika, “Al Qaeda in the Islamic Maghreb
(AQIM) and the Alleged Production of the Etiological Agent of Plague,” ASA Newsletter 131 (April 2009): p. 1, 21-22,
http://www.asanltr.com/newsletter/09-2/articles/092a.pdf.
2506 For the accounts themselves, see:
Eli Lake, “Al Qaeda bungles arms experiment,” The Washington Times, January 19, 2009,
http://www.washingtontimes.com/news/2009/jan/19/al-qaeda-bungles-arms-experiment/
2507 Olivier Guitta, “Al-Qaeda in the Islamic Maghreb: A Threat for the West,” Defence Against Terrorism Review 3, no. 1
(Spring 2010): p. 57-58, http://www.coedat.nato.int/publication/datr/volume5/03-Al-
Qaeda_in_the_Islamic_Maghreb_A_Threat_for_the_West.pdf.
2508 Mike M. Ahlers, Brian Todd, “Al Qaeda group contemplated poisoning food in U.S., officials say,” December 22, 2010,
http://www.cnn.com/2010/US/12/21/al.qaeda.poison.plot/.

Islamic Movement of
NO NO
Uzbekistan (IMU)
Real Irish Republican
NO NO
Army (RIRA)
Jaish-e-Mohammed
NO NO
(JEM)
Lashkar-e Tayyiba (LeT) NO NO
The group claimed to possess
chemical and biological weapons in
Al-Aqsa Martyrs
YES NO 2006 in an attempt to deter Israeli
Brigade (AAMB)
military action.2509 This claim is
regarded as spurious.2510
Asbat al-Ansar (AAA) NO NO
NO, albeit Recent Philippines military claim
Communist Party of the
reported war that NPA used feces to spike
Philippines/New
NO use of explosive devices to cause sepsis;
People's Army
biological see FARC and ELN entries.2511 The
(CPP/NPA)
material NPA denies this.
Attempted production of BW,
mostly as part of Al Qaeda’s
YES program, with unknown results.
Jemaah Islamiya (JI) (By proxy YES NO See detailed entry. Group
with AQ) membership, including leadership
and individuals involved in the BW
program, decimated.
Pakistani police reportedly
Lashkar i Jhangvi (LJ) NO NO uncovered chemical laboratories
belonging to the group.2512
Gary Ackerman, Jeremy Tamsett (Boca Raton: CRC Press, 2009), p. 143.
2510 Ibid.
http://mindanaoexaminer.com/philippine-army-finds-human-feces-snake-venom-in-wounded-soldiers-wounds/.
2512 National Consortium for the Study of Terrorism and Responses to Terrorism (START), “Terrorist Organization Profiles:
Lashkar-e-Jhangvei,” http://www.start.umd.edu/tops/terrorist_organization_profile.asp?id=65.

Initial reports held that the group
had a poison laboratory in Iraq that
Unsubstantiat manufactured botulinum and ricin
ed reports of toxin.2513 However, in retrospect,
Ansar al-Islam (AAI) NO NO
interest in the site does not appear to have
toxins produced toxins. The site is not
mentioned in the report of the Iraq
Survey Group.2514
Continuity Irish
Republican Army NO NO
(CIRA)
Libyan Islamic Fighting
NO NO
Group (LIFG)
2513 Preston Mendenhall, “Positive test for terror toxins in Iraq,” MSNBC.com, April 4, 2003,
http://www.nbcnews.com/id/3070394/ns/world_news/t/positive-test-terror-toxins-iraq/#.VXdWckbrJ-A.
2514 Milton Leitenberg, Assessing the Biological Weapons and Bioterrorism Threat, Strategic Studies Institute monograph,
December 2005, p. 26-27, http://www.strategicstudiesinstitute.army.mil/pdffiles/pub639.pdf.

Emerging group with enormous
resources. Reports of chemical
munitions use (chlorine, phosphine,
and mustard).2515,2516,2517,2518,2519
One individual member had a
laptop with over 35,000 files
Islamic State of Iraq and dedicated to Jihad, a few of which
the Levant (formerly al- NO Unknown discussed BW.2520 Concern over
Qa'ida in Iraq) alleged looting of biological
laboratories in Syria.2521 In 2014,
DHS secretary Jeh Johnson stated
that his service had “seen no
specific credible intelligence that
ISIS is attempting to use any sort of
disease or virus to attack our
homeland.”2522
Islamic Jihad Union
NO NO
(IJU)
Harakat ul-Jihad-i-
Islami/Bangladesh NO NO
(HUJI-B)
al-Shabaab NO NO
Revolutionary Struggle
NO NO
(RS)
Kata'ib Hizballah (KH) NO NO
Harakat ul-Jihad-i-Islami
NO NO
(HUJI)
2515 Tom Coghlan, Catherine Philp, Ammar Shamary, “Jihadists unleash chemical weapons in battle for Tikrit,” The Times,
March 14, 2015, <http://www.thetimes.co.uk/tto/news/world/middleeast/article4381521.ece>.
2516 “Chlorine bomb attacks by jihadists are growing threat to the UK, warns chemical warfare expert,” The Independent, May
25, 2015, http://www.independent.co.uk/news/uk/home-news/chlorine-bomb-attacks-by-jihadists-are-growing-threat-to-the-
uk-warns-chemical-warfare-expert-10274947.html.
2517 Phosphine, chemical formula PH3, is used as a fumigant, but is toxic if inhaled. Ibid; also see: Sajila Saseendran, “Ministry
mulls banning `killer’ pesticide,” Khaleej Times, September 2, 2014,
http://www.khaleejtimes.com/article/20140901/ARTICLE/309019899/1002.
2518 Nabih Bulos, “Islamic State confirmed to have used mustard gas against Kurds in Syria,” The Telegraph, August 15, 2015,
http://www.telegraph.co.uk/news/worldnews/middleeast/syria/11805235/Islamic-State-confirmed-to-have-used-mustard-
gas-against-Kurds-in-Syria.html.
2519 Paul Blake, “US official: ÌS making and using chemical weapons in Iraq and Syria’,” BBC News, September 11, 2015,
http://www.bbc.com/news/world-us-canada-34211838.
http://foreignpolicy.com/2014/08/28/found-the-islamic-states-terror-laptop-of-doom/.
2521 Ari Soffer, “Experts Warn of Al Qaeda Biological Weapons Threat,” Israel National News, October 16, 2013,
http://www.israelnationalnews.com/News/News.aspx/172897#.VXddiUbrJ-A.
2522 “Use of Ebola virus as bioterror weapon highly unlikely: Experts,” Homeland Security News Wire, November 11, 2014,
http://www.homelandsecuritynewswire.com/dr20141111-use-of-ebola-virus-as-bioterror-weapon-highly-unlikely-experts.

Tehrik-e Taliban
NO NO
Pakistan (TTP)
Jundallah NO NO
Army of Islam (AOI) NO NO
A 2010 email claiming to be from
“Indian Mujahedeen (Assam)”
Indian Mujahedeen (IM) YES NO
threatened biological warfare
unless its demands were met.2523
Jemaah Anshorut Tauhid Splinter group from Jemaah
NO NO
(JAT) Islamiyah.
Abdallah Azzam
NO NO
Brigades (AAB)
Haqqani Network
(HQN)
Ansar al-Dine (AAD) NO NO
Boko Haram NO NO
Ansaru NO NO
al-Mulathamun Battalion NO NO
Ansar al-Shari'a in
NO NO
Benghazi
Ansar al-Shari'a in
NO NO
Darnah
Ansar al-Shari'a in
NO NO
Tunisia
Ansar Bayt al-Maqdis NO NO
Emerging group. Concern over
al-Nusrah Front Unknown Unknown alleged looting of biological
laboratories in Syria.2524
Mujahidin Shura
Council in the Environs NO NO
of Jerusalem (MSC)
http://www.nti.org/gsn/article/extremists-warn-of-biological-strike-in-india/.
2524 Ari Soffer, “Experts Warn of Al Qaeda Biological Weapons Threat.”

16.10 Detailed History of Known Terrorist Biological Weapons Programs
Based on the research presented below, three international terrorist groups (Al Qaeda, Jemaah
Islamiyah, and Aum Shinrikyo) and two domestic terrorist groups (Rajneesh Cult, R.I.S.E.), have sought
a biological weapons capability intended for mass casualty attacks. Only the Rajneesh Cult launched a
successful, albeit rudimentary, biological weapons attack. The Rashneeshees hoped to sicken, but did not
seek to kill, many individuals. Aum Shinrikyo wished to cause deaths, but failed in all of its biological
weapons attack attempts. Of the five groups, only Al Qaeda is likely to be pursuing a biological weapons
capability at the present time. Jemaah Islamiyah’s membership, including its core leadership and all
known BW-program members, has been decimated in recent years due to a string of arrests. Aum
Shinrikyo’s WMD program has been dismantled, and key members are in jail. As for R.I.S.E., it is a long-
defunct group that only had two core members.
Finally, one apparent “false positive” case exists in older reports in which a group that initially was
flagged as had interest in BW appear to have been incorrect, or at least remain unsubstantiated by publicly
available information. However, open source literature references that the Weather Underground group
(disbanded in 1976) had attempted to blackmail a homosexual officer to obtain incapacitating agents from
the U.S. Army’s Fort Detrick BW research center, with the ultimate goal being to incapacitate (but not
kill) individuals by poisoning a U.S. city water supply.2525 A thorough review of this case by John V.
Parachini in a volume edited by Jonathan B. Tucker concludes: “contrary to the conventional wisdom, the
Weather Underground probably did not seek to acquire or employ biological or chemical weapons.”2526
Summarizing these case studies, only a select few terrorist groups have demonstrated their intent to use
biological weapons to cause mass casualties in the past. These programs involved few core members
(from two to fourteen), and engaged even fewer individuals with advanced training in the life sciences.
All four groups attempted to obtain or obtained virulent pathogenic strains by acquiring seed cultures
from laboratories by leveraging some form of insider access. Of these groups, only Al Qaeda’s efforts in
acquiring BW are thought to be ongoing.
The following vignettes review the aforementioned five confirmed bioterrorist programs:
16.10.1 Al Qaeda
Al Qaeda has worked on biological weapons with the intent to cause mass casualties since before the 9/11
attacks. Significant uncertainties remain as to the group’s achievements.
16.10.1.1 Motivation and intent to use
Official Al Qaeda statements underline the group’s intent to use biological weapons against U.S. citizens.
An internal account of discussions occurring within the group’s ruling body, the Majlis al-Shura, makes
clear that in 1998, the leadership debated the utility of pursuing WMD.2527 The supporters of WMDs in
that debate won out, apparently through rhetoric that a WMD capability would be needed to prevent
America and Israel from employing WMDs; in essence, they argued that WMD was necessary for
deterrence.2528 Bin Laden announced his strong support for WMD in a December 22, 1998 interview,
which specifically mentioned biological weapons:
2525 John V. Parachini, “The Weather Underground (1970),” Toxic Terror: Assessing Terrorist Use of Chemical and Biological
Weapons, ed. Jonathan B. Tucker (Cambridge: Belfer Center for Science and International Affairs, 2000), p. 43, 47.
2526 Ibid.
2527 Sammy Salama, Lydia Hansell, “Does Intent Equal Capability? Al-Qaeda and Weapons of Mass Destruction,”
Nonproliferation Review 12, no. 3 (November 2005): p. 625-626, p. 650fn.84.
2528 Ibid.

“Muslim scholars have issued a fatwa against any American who pays taxes to his government.
He is our target, because he is helping the American war machine against the Muslim
nation.”2529
Bin Laden then stated, specifically mentioning biological weapons:
“It is very strange: if America has all the mass-destruction weapons, that is nothing. If the
Jewish state has the same weapons, that is OK. But if a Muslim state like Pakistan tries to
defend itself against the Hindu hegemony in South Asia, everything should be done to prevent it
from doing so. We don’t consider it a crime if we tried to have nuclear, chemical, biological
weapons. Our holy land is occupied by Israeli and American forces. We have the right to defend
ourselves and to liberate our holy land.”2530
After the 9/11 attacks and the expulsion of Al Qaeda from Afghanistan, the group’s rhetoric embraced the
notion of first strike. On June 12, 2002, Al Qaeda spokesman Abu Ghaith produced a three-part article
titled “In the Shadow of the Lances” that included the following passage:
“We have the right to kill four million Americans – two million of them children – and to exile
twice as many and wound and cripple hundreds of thousands. Furthermore, it is our right to
fight them with chemical and biological weapons, so as to afflict them with the fatal maladies
that have afflicted the Muslims because of the [Americans’] chemical and biological
weapons.”2531
On May 21, 2003, radical cleric Nasir al Fahd issued a fatwa permitting the use of WMDs just before his
arrest.2532 In “A Treatise on the Legal Status of Using Weapons of Mass Destruction against Infidels,"
Nair al Fahd attempted to justify: 1) the use of techniques that would not kill individuals “in a good
manner,” such as acts to “bomb, destroy, burn or flood”; 2) the killing of children and women; and 3) the
killing of Muslims.2533,2534 Nasir al Fahd’s statement was harnessed by Al Qaeda’s ideologues. For
instance, he is cited by Ayman Zawahiri, the current leader of Al Qaeda, in the latter’s 2008 book
Exoneration.2535
Al Qaeda’s Encyclopedia of Jihad, an eleven volume training manual, reportedly includes instructions for
the manufacture and use of chemical and biological weapons; if true, the topic’s inclusion in the manual
reinforces the evidence that the group sees the use of these weapons as justified and desirable.2536 A 2012
article published by the Yemen branch of Al-Qaeda (AQAP) in the group’s jihadist magazine Inspire
confirms that this interpretation remains current, since it emphasized that “the use of poisons of chemical
2529 Jamal Ismail, “I Am Not Afraid of Death,” Newsweek, January 10, 1999, http://www.newsweek.com/i-am-not-afraid-death-
165374.
2530 Ibid.
2531 The Middle East Media Research Institute (MEMRI), “Contemporary Islamist Ideology Authorizing Genocidal Murder,”
Special Report No.25, January 27, 2004, http://www.memri.org/report/en/0/0/0/0/0/0/1049.htm#_ednref21.
2532 Rolf Mowatt-Larssen, “Islam and the Bomb: Religious Justification For and Against Nuclear Weapons,” Working paper,
Harvard Kennedy School Belfer Center for Science and International Affairs, January 2011, p. 38,
http://belfercenter.hks.harvard.edu/files/uploads/Islam_and_the_Bomb-Final.pdf.
2533 Ibid.
2534 Sammy Salama, Edith Bursac, “Jihadist Capabilities and the Diffusion of Knowledge,” Jihadists and Weapons of Mass
Destruction, eds. Gary Ackerman, Jeremy Tamsett (Boca Raton: CRC Press, 2009).
2535 International Research Center (IRC), “Zawahiri Tries to Clear Name, Explain Strategy,” Transnational Security Issues
Report, April 21, 2008, p.4, retrieved at: https://fas.org/irp/eprint/zawahiri.pdf.
2536 Sammy Salama, Lydia Hansell, “Does Intent Equal Capability? Al-Qaeda and Weapons of Mass Destruction,” p. 618;
Nick Fielding, “Encyclopedia of Terror: Revealed: The bloody pages of Al-Qaeda’s killing manual,” The Sunday Times of
London, April 11, 2001.

and biological weapons against population centers is allowed and strongly recommended due to the effect
on the enemy.”2537
16.10.1.2 Program history
A former senior CIA official, Rolf Mowatt-Larssen, who is publicly described as having “led the U.S.
government’s efforts to determine whether al Qaeda had WMD capabilities,” provides several key details
in the following account of Al Qaeda’s BW efforts.2538 Wherever possible, open source reporting has been
used to corroborate, expand upon, or challenge his account.
A systematic pre-9/11 Al Qaeda biological weapons program was launched and headed by Ayman
Zawahiri, following the merger of the latter’s Egyptian Islamic Jihad group with Al Qaeda Central in
early 1998.2539 Ayman Zawahiri was a long-time proponent of biological weapons, and is today the leader
of Al Qaeda Central.
More specifically, Zawahiri communicated electronically with Mohammed Atif, the head of Al Qaeda’s
Military Committee (killed in 2001), on the setup of a chemical and biological weapons project named
“al-Zabadi.”2540 A message by Zawahiri to Atif dated April 15, 1999 pushed for the development of
biological weapons, arguing that “the destructive power of these weapons is no less than that of nuclear
weapons” in terms of causing mass casualties.2541 The message made clear that Zawahiri was looking to
hire an expert: “I would like to emphasize what we previously discussed—that looking for a specialist is
the fastest, safest, and cheapest way. Simultaneously, we should conduct a search on our own.”2542 The
proposed start-up budget was outlined as $2000-4000 USD.2543
Zawahiri proceeded with this plan and recruited Rauf Ahmed, a mid-level Pakistani government biologist,
into Al Qaeda’s biological weapons efforts that same year.2544 Ahmed’s tasking included the setup of a
laboratory in Kandahar, Afghanistan.2545 Rauf Ahmed corresponded with Zawahiri, initially noting
trouble with finding a B. anthracis strain. One letter read, verbatim: “Unfortunately, I did not find the
required culture of B. anthrax – i.e., pathogenic.”2546 Whether Rauf Ahmed ever managed to obtain a
pathogenic strain is not known based only on open source information.2547
The following passage from historian Christopher Andrew, based on internal MI5 documents made
available to the researcher, fill in some gaps:
2537 Gary A. Ackerman, Lauren E. Pinson, “An Army of One: Assessing CBRN Pursuit and Use by Lone Wolves and
Autonomous Cells,” Terrorism and Political Violence 26, no. 1 (2014): p. 229-230;
“Al-Qaeda Magazine Urges Chemical, Biological Strikes Against Foes,” NTI Global Security Newswire, May 3, 2012,
http://www.nti.org/gsn/article/al-qaeda-magazine-urges-chemical-biological-strikes-us/.
2538 Rolf Mowatt-Larssen, “Al Qaeda’s Pursuit of Weapons of Mass Destruction,” Foreign Policy, January 25, 2010,
http://foreignpolicy.com/2010/01/25/al-qaedas-pursuit-of-weapons-of-mass-destruction/.
2539 Ibid.
2540 Alan Cullison, Andrew Higgings, “Computer in Kabul holds chilling memos,” The Wall Street Journal, December 31, 2001.
2541 Alan Cullison, “Inside Al-Qaeda’s Hard Drive,” The Atlantic, September 2004,
http://www.theatlantic.com/magazine/archive/2004/09/inside-al-qaeda-s-hard-drive/303428/.
2542 Ibid.
2543 Ibid.
2544 Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?,” p. 14.
2545 Ibid.
2546 Joby Warrick, “Suspect and A Setback In Al-Qaeda Anthrax Case,” The Washington Post, October 31, 2006, p.3,
http://www.washingtonpost.com/wp-dyn/content/article/2006/10/30/AR2006103001250_3.html.
2547 Whether Al Qaeda in general ever obtained a pathogenic strain of B. anthracis has not been disclosed in open sources, as
discussed below in the current document.
Judith Miller, “U.S. Has New Concerns About Anthrax Readiness,” The New York Times, December 28, 2003,
http://www.nytimes.com/2003/12/28/national/28ANTH.html?pagewanted=1.

“In September 2000 the Pakistani microbiologist Rauf Ahmad attended a conference in Britain
on dangerous pathogens, where he sought samples from other delegates as well as help in
obtaining a bioreactor and cell counter.[2548] The Service [MI5] was alerted to his activities and
a search of his luggage on departure from the UK revealed £13,000, which he claimed was ‘to
buy equipment’, documents detailing his contacts (including UK companies) and a copy of his
CV. The CV revealed that Ahmad had a PhD from a university in Pakistan, had attended earlier
conferences in Britain in 1997 and 1999 and had published scientific papers on anthrax.
Security Service officers visited the UK companies with which Ahmad had made contact and
they broke off their dealings with him. Ahmad’s visits to Britain had much greater significance
than was apparent at the time. Their purpose only became clear after 9/11, from documents
recovered by US forces in Afghanistan in 2001. Among the documents was correspondence
between ‘Abu Mohamed’ and ‘Abu Ibrahim’ about procurement of equipment, cultures and
training for BW production. ‘Abu Mohamed’ was quickly identified as UBL’s [Osama bin
Laden’s] deputy, Ayman al Zawahiri. ‘Abu Ibrahim’ took longer to track down. References in
the correspondence to his foreign travels, attendance at conferences in the UK and attempts to
procure dangerous pathogens, however, were discovered to match exactly the information on
Ahmad in Security Service files.”2549
Yazid Sufaat was recruited to work on the Al Qaeda program after Rauf Ahmed was hired. Accounts vary
as to exactly when and why this occurred, but Sufaat was brought in no later than early 2001.2550,2551,2552
The leader of Jemaah Islamiyah, an Al Qaeda-allied group based in Indonesia, presented Yazid Sufaat to
Zawahiri.2553 Yazid Sufaat was an ex-Malaysian Army captain with a biochemistry degree from a U.S.
university (California State University-Sacramento).2554 Following this introduction, Zawahiri entrusted
Sufaat with acquiring and preparing a sample of B. anthracis for production.2555 Sufaat embarked on this
work at a hospital laboratory in Kandahar.2556 Zawahiri kept Ahmed’s and Sufaat’s endeavors
compartmentalized and neither knew of the other’s existence.2557 This was either good tradecraft, as Rolf
Mowatt-Larssen alleges, or simply the result of having fired Rauf over the latter’s constant requests for
money and dubious loyalty to the group before hiring Sufaat, as most other sources hold.2558 One month
before the 9/11 attacks in 2001, Ayman Zawahiri inspected Ahmed’s laboratory.2559 Zawahiri was also
2548 Rauf Ahmad is sometimes used instead of Rauf Ahmed in other accounts. See for example: George Tenet, At the Center of
the Storm: My Years at the CIA (New York: HarperTorch, 2007), p. 278.
2549 Christopher Andrew, Defend the Realm: The Authorized History of MI5 (New York: Vintage Books, 2009), p. 807-808.
2550 Rolf Mowatt-Larssen’s account says this occurred in early 1999, as part of the setup of a “second, parallel network.” René
Pita and Rohan Gunaratna, who interviewed intelligence service personnel who arrested and interrogated Ahmed, state this
occurred in 2000, when Zawahiri grew dissatisfied with Ahmed’s work. Finally, the 9/11 Report states that “Sufaat did not
start on the al Qaeda biological weapons program until after JI’s December 2000 church bombings in Indonesia, in which he
was involved.” Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?,”
2551 René Pita, Rohan Gunaratna, “Revisiting Al-Qaìda’s Anthrax Program,” CTC Sentinel Vol. 2 Issue 5, May 2009, p. 2,
p.2fn.21, <https://www.ctc.usma.edu/posts/revisiting-al-qaida%E2%80%99s-anthrax-program>
2552 The National Commission on Terrorist Attacks Upon the United States, The 9/11 Commission Report, p.490fn.23,
http://www.9-11commission.gov/report/.
2553 Rolf Mowatt-Larssen, “Al Qaeda’s Pursuit of Weapons of Mass Destruction.”
2556 Rolf Mowatt-Larssen, “How to Get Terrorists to Talk,” The National Interest, February 18, 2015, p.2,
http://nationalinterest.org/feature/how-get-terrorists-talk-12270?page=2.
2558 See footnote 21. Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?,” p. 14-15;
René Pita, Rohan Gunaratna, “Revisiting Al-Qaìda’s Anthrax Program,” CTC Sentinel Vol. 2 Issue 5, May 2009, p. 2,
p.2fn.21, https://www.ctc.usma.edu/posts/revisiting-al-qaida%E2%80%99s-anthrax-program.

briefed by Sufaat on the latter’s efforts on cultivating a pathogenic B. anthracis strain.2560
The subsequent U.S. outing of the Taliban in Afghanistan disrupted the efforts described above. Ahmed
was detained in Pakistan and Sufaat was arrested in Malaysia in December 2001.2561 Pakistan cut off
FBI’s access to Rauf Ahmed in 2003, and he is now free.2562
Sensitive site exploitation of Al Qaeda camps in Afghanistan following the ousting of the Taliban
unearthed evidence of the group’s biological weapons program. Ahmed’s laboratory was uncovered; Rolf
Mowatt-Larssen’s 2015 account describes it as having been “crude” and used to store purchased
equipment.2563 A U.S. defense department spokesperson briefed the press on September 14, 2002 with
photographs showing a centrifuge for liquid separation and a dryer said to have been discovered at a BW
laboratory in Kandahar; although publicly unconfirmed, based on the description given this was probably
Ahmed’s.2564 Moreover, one Al Qaeda site in Afghanistan whose name and location has not been
disclosed held over 20 old research articles from U.K. journals that together “provided a method for
isolating, culturing, identifying, and producing bacteria, including Bacillus anthracis and Clostridium
botulinum.”2565
Exploitation of Al Qaeda camps in Afghanistan also revealed some degree of operative training in
biological warfare. Two training camps, the Durante and Tarnak Farms, reportedly provided basic
training to operatives on biological weapons matters; however, since details have not been made public,
these allegations may be restricted to toxins.2566 These camps were run by chemist Abu Khabab al-Masri
and by Abu Musab al-Suri.2567 Both were proponents of the use of weapons of mass destruction (WMD)
against the United States.2568 Training manuals written by Abu Khabab al-Masri “that contain instructions
for making chemical and biological weapons […] were recovered by U.S. forces in Afghanistan.”2569 Abu
Musab al-Suri was captured in 2005 and Abu Khabab al-Masri was killed in 2008.2570
16.10.1.3 Capability assessment
Whether any B. anthracis was produced by Al Qaeda, and whether Al Qaeda obtained the necessary B.
anthracis seed cultures to do so, remains unclear. One individual that CIA believed was involved in Al
Qaeda’s BW program was captured and subjected to what the Senate Select Committee on Intelligence
2560 Ibid.
2561 Joby Warrick, “Suspect and A Setback in Al-Qaeda Anthrax Case,” The Washington Post, October 31, 2006,
Maria Ressa, “Reports: Al Qaeda operative sought anthrax.”
2562 Joby Warrick, “Suspect and a Setback in Al-Qaeda Anthrax Case.”
2563 Rolf Mowatt-Larssen, “How to Get Terrorists to Talk,” p.2.
2564 The equipment pictured could not be independently assessed in the present report, because as the media article cited below
notes: “The Department of Defense refused to make available the photos of the dryer and the centrifuge it said came from
the lab, or any of the other photos and slides discussed at today's briefing. In response to a reporter's question, the senior
official said the department had arranged the briefing in response to reporters' requests for an unclassified version of the
secret briefing on these subjects that Mr. Rumsfeld had been giving.” In: Judith Miller, “Lab Suggests Qaeda Planned to
Build Arms, Officials Say,” The New York Times, September 14, 2002,
http://www.nytimes.com/2002/09/14/international/asia/14LAB.html.
2565 James B. Petro, David A. Relman, “Understanding Threats to Scientific Openness,” Science 302, no. 5652 (December 12,
2003): p. 1898, http://www.sciencemag.org/content/302/5652/1898/suppl/DC1.
2566 Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?” p. 13-14.
2567 Ibid.
2568 Ibid.
2569 Ibid.
2570 “Al Qaeda: Weapons expert among dead `heroes’,” CNN, August 3, 2008,
http://www.cnn.com/2008/WORLD/asiapcf/08/03/terrorist.killed/.

has described as “harsh treatment” during the interrogation that followed.2571 He initially said, “we never
made anthrax.”2572 Once he was told that the interrogation would not stop until “he told the truth,” he then
stated, crying, “I made the anthrax.”2573 Prompted, he then said he was lying.2574 Interrogators
“demonstrated the penalty for lying.”2575 The individual then repeated the “I made the anthrax” claim,
promptly recanted the statement, and finally re-stated the production claim.2576 In questioning two days
later, the individual stated that he had lied “only because he thought that that was what interrogators
wanted.”2577 News reports on the seizure of the laboratory facility failed to clarify whether the laboratory
had been used to produce biological agents.2578 Finally, the National Research Council (NRC) of the U.S.
National Academies, in their “Review of the Scientific Approaches Used During the FBI’s Investigation
of the 2001 Bacillus anthracis mailings,” remarked:
“At the end of this study, the committee was provided limited information for the first time
about the analysis of environmental samples for B. anthracis Ames from an undisclosed
overseas site at which a terrorist group’s anthrax program was allegedly located. This site was
investigated by the FBI and other federal partners as part of the anthrax letters investigation.
The information indicates that there was inconsistent evidence of Ames strain DNA in some of
these samples, but no culturable B. anthracis.”2579
According to Rolf Mowatt-Larssen, the FBI took samples from Sufaat’s hospital laboratory.2580 However,
the link between this sampling activity and the operation mentioned in the NRC report has not been
confirmed in open sources. Since the analysis of any gathered samples has not been made public, no open
source evidence is available to assess whether Al Qaeda had successfully isolated a pathogenic strain of
B. anthracis.
Little is known regarding any Al Qaeda plans for the B. anthracis once produced. However, indicators of
Al Qaeda interest in crop duster airplanes of uncertain reliability existed.2581,2582,2583 Large-scale aerosol
2571 Senate Select Committee on Intelligence, Committee Study of the Central Intelligence Agency’s Detention and Interrogation
Program, Foreword by Senate Select Committee on Intelligence Chairman Dianne Feinstein, Findings and Conclusions,
Executive Summary, unclassified, approved December 13, 2012, updated for release April 3, 2014, declassification
revisions December 3, 2014, p. 82fn.442, http://www.intelligence.senate.gov/sites/default/files/publications/CRPT-
113srpt288.pdf.
2572 Ibid.
2573 Ibid.
2574 Ibid.
2575 Ibid.
2576 Ibid.
2577 Ibid.
2578 Compare: “The lab had been abandoned by Al Qaeda before production began, officials said.” In: Judith Miller, “Lab
Suggests Qaeda Planned to Build Arms, Officials Say.”
With: “U.S. officials said the evidence neither establishes nor rules out that al Qaeda completed manufacture.” In: Barton
Gellman, “Al Qaeda Near Biological, Chemical Arms Production,” Washington Post, March 23, 2003,
2579 National Research Council of the National Academies, Review of the Scientific Approaches Used During the FBI’s
Investigation of the 2001 Bacillus Anthracis Mailings (Washington: The National Academies Press, 2011), p. 8, 72.
2580 Rolf Mowatt-Larssen, “How to Get Terrorists to Talk,” p.2.
2581 Sammy Salama, Lydia Hansell, “Does Intent Equal Capability? Al-Qaeda and Weapons of Mass Destruction,”
2582 Julian Borger, “Cropdusters grounded in poison alert,” The Guardian, September 23, 2001,
http://www.theguardian.com/world/2001/sep/24/afghanistan.terrorism9
2583 “The FBI reviewed a list of some 11,000 agricultural aircraft provided by the Federal Aviation Administration, according to
documents provided to the Sept. 11 commission. Working from that list, agents interviewed and did background checks on
3,028 operators and owners of the planes.” In:
“FBI Checking Crop-Dusting Planes and Pilots, Still Worried About Possible Terror Use,” Police One, April 22, 2004,
http://www.policeone.com/terrorism/articles/85144-FBI-Checking-Crop-Dusting-Planes-and-Pilots-Still-Worried-About-
Possible-Terror-Use/.

dissemination may therefore have been a possibility, a threat which FBI took extremely seriously.2584
Few details have surfaced regarding Al Qaeda’s post-2001 BW efforts. One exception was a document
titled “Terror Franchise: The Unstoppable Assassin, TECHS Vital role for its success,” found during the
2011 raid that killed Bin Laden. It was prepared by a senior Al Qaeda member, Abu-Salih al Somali, and
must have been written sometime after 2009 given the events referenced in the text.2585
This document places particular emphasis on the use of poisons and biological weapons, which it classes
as “toxicants.” The document, written in broken English, is addressed to “Engineers, Doctors, Biologists,
Pharmacists, researchers, hobbyists, Handymen and women, experimenters, discoverers, The courageous,
Experts in all fields, Amateurs, and all of you who care and realize that you are part of an Ummah.”2586 It
calls for assistance from these “techs” in producing and disseminating knowledge and know-how on,
inter-alia, “how to be able to make death, in its explosions form- especially (the Oxidizer part of it) and
toxicants in an easy, practical and improvised way anywhere on earth...” [Emphasis and punctuation in
original]2587
The document reminds the reader that “Americans and their NATO allies’ citizens” are to be targeted in a
campaign to cause “nonstop, unpredicted, invisible sudden death,” and gives as an example the use of
cyanide or ricin on products sold by supermarkets and restaurants.2588 The document ends with a detailed
list of military topics the author is requesting the “techs” to research and subsequently share their findings
through instruction manuals and videos. The following list of requests, emphasized as “immediately
needed,” is found under the “toxicants” request section:
1. Actual improvised production and testing of Cyanides, Ricin.(immediately needed),
2. Preparation and testing (rabbits is ok) of any lethal (delayed and immediate) ingested toxicants,
3. On Camera production of any of war gases (Phosgene, VX, etc)..Look at Nbk file and scientific
principles of improvised home warfare. (onsite production apparatus also),
4. Actual production and testing of Biological toxicants (Anthrax, Botulinum, clostridium,

endotoxins, Exotoxins, etc),
5. Production of (HCl) or whatever is needed in the production of toxicants,
6. Any other options that can be used as toxicant… plants, etc… detailed, local names pictures,
incidents, cultivation…Insects…etc… read scientific principles of improvised home warfare
volume 2,5,6,
7. Bacteria based weapons…how? detail. Any other practical options,

2585 David Francis, “Al Qaeda’s Blueprint For How To Start a Homegrown Terror Franchise,” Foreign Policy, May 20, 2015,
http://foreignpolicy.com/2015/05/20/al-qaedas-blueprint-for-how-to-start-a-homegrown-terror-franchise/.
bookshelf?start=1;
Retrieved under the “Now Declassified Material” folder: Abu-Salih Al Somali, “Terror Franchise: The Unstoppable
Assassin, TECHS Vital role for its success,” http://www.dni.gov/files/documents/ubl/english/Terror%20Franchise.pdf.
2587 Ibid.
2588 Ibid.

8. Airborne substance that when sprayed in small quantity or mixed (tablet form) with water,
tranquilizes the entire inhabitants of a hall or plane. And its antidote.2589
This document confirms the group’s continued interest in obtaining and using biological weapons to
cause mass civilian deaths. At the same time, it highlights the group’s paucity of expertise in the matter,
at least as of 2009.
16.10.2 Jemaah Islamiyah
Jemaah Islamiyah is a Southeast Asian terrorist group that has been allied with Al Qaeda since 1998.2590
The group has been in sharp decline, although it remains on the U.S. list of Foreign Terrorist
Organizations.2591,2592,2593 Jemaah Islamiyah had a joint biological warfare program with Al Qaeda
organized by Riduan Isamuddin and run by Sufaat. Riduan Isamuddin was Jemaah Islamiyah’s director of
operations.2594 He oversaw the group’s financing, led Jemaah Islamiyah’s regional policy-making organ,
and organized Al Qaeda’s regional operations.2595 Riduan Isamuddin suggested and organized the transfer
of Sufaat to Al Qaeda’s BW program. When U.S. operations in Afghanistan began in October 2001,
Sufaat fled Afghanistan for Bogor, Indonesia.2596 He then sought to set up a new BW program in-country,
but failed to recruit a microbiologist at an Indonesian institute.2597,2598,2599 He was captured in December
2001; Isamuddin was apprehended in 2003.2600 A Jemaah Islamiyah manual indicating interest in
chemical and biological weapons was reportedly discovered in the Philippines in 2003.2601
16.10.3 Aum Shinrikyo
The Japan-based millenarian cult and terrorist group, Aum Shinrikyo, embarked on a WMD program in
1990, intending to cause mass casualties and precipitate the apocalypse.2602 The group’s WMD network
was dismantled following their March 1995 sarin nerve agent attack in the Tokyo Subway and the
subsequent arrest of top Aum leaders. However, Aum Shinrikyo remains on the U.S. list of Foreign
2589 Ibid.
2590 The National Commission on Terrorist Attacks Upon the United States, The 9/11 Commission Report, p. 151.
2591 Fouad Pervez, “Jemaah Islamiyah,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A.
Zilinskas (Hoboken: John Wiley & Sons, 2011), p. 370.
2592 National Counterterrorism Center, “Jemaah Islamiyah (JI),” September 2013, http://www.nctc.gov/site/groups/ji.html.
2593 U.S. Department of State, “Foreign Terrorist Organizations,” <http://www.state.gov/j/ct/rls/other/des/123085.htm
2594 United Nations Security Council Committee pursuant to resolutions 1267 (1999) and 1989 (2011) concerning Al-Qaida and
associated individuals and entities, “QDi.087 Nurjaman Riduan Isamuddin,” March 28, 2011,
http://www.un.org/sc/committees/1267/NSQDi087E.shtml.
2595 Ibid.
2596 Judith Miller, “U.S. Has New Concerns About Anthrax Readiness,” The New York Times, December 28, 2003,
http://www.nytimes.com/2003/12/28/us/us-has-new-concerns-about-anthrax-readiness.html.
2597 Ibid.
2599 René Pita, Rohan Gunaratna, “Revisiting Al-Qaìda’s Anthrax Program,”
2600 Rolf Mowatt-Larssen, “Al Qaeda Weapons of Mass Destruction Threat: Hype or Reality?” p. 28.
2601 Christopher Torchia, “Experts: Bioterrorism Should Worry Asia,” Associated Press, March 25, 2006.
edition,” Center for a New American Security, December 2012, p. 18-20,

Terrorist Organizations, perhaps in part because of concerning reports in 2000 that the cult was
regrouping.2603,2604,2605
The group’s biological weapons program was based on B. anthracis, alongside a toxin weapons program
focused on botulinum toxin and a chemical weapons program focused primarily but not exclusively on
the nerve agent sarin.2606 Certain Aum Shinrikyo members voiced a passing interest in Ebola virus as a
weapon, although no evidence that the group attempted to acquire a pathogen sample exists.2607,2608
The group’s biological weapons team was composed of about ten individuals, led by a graduate-trained
molecular biologist named Seiichi Endo.2609 Endo had taken courses in molecular biology and genetic
engineering at the PhD level at the Viral Research Center at Kyoto University, but did not complete
enough coursework to obtain a doctorate degree.2610 This team drew upon cult rank-and-file members to
carry out equipment purchases. The team’s BW endeavor was sustained by the group’s significant
infrastructure and finances, which allowed for the liberal purchase of laboratory equipment and the
acquisition of reference texts.2611 The cult specifically targeted scientists, engineers, and technicians that
could be of use for the group’s weapons programs in their recruitment campaigns.2612
After an abortive phone call to the U.S. Centers for Disease Control and Prevention, the group decided
against attempting to purchase a strain from a U.S. culture collection for fear of being discovered.2613
Endo reportedly attempted and failed to steal a B. anthracis strain from a laboratory, after which an
2603 U.S. Department of State, “Foreign Terrorist Organizations,” http://www.state.gov/j/ct/rls/other/des/123085.htm.

2604 David E. Kaplan, “Aum Shinrikyo,” Toxic Terror: Assessing Terrorist Use of Chemical and Biological Weapons, ed.
Jonathan B. Tucker (Cambridge:Belfer Center for Science and International Affairs.
2605 Sara Daly, John Parachini, William Rosenau, Aum Shinrikyo, Al Qaeda, and the Kinshasa Reactor: Implications of Three
Case Studies for Combating Nuclear Terrorism (Santa Monica: RAND, 2005), p. 12,
http://www.rand.org/content/dam/rand/pubs/documented_briefings/2005/RAND_DB458.pdf.
2606 Raymond A. Zilinskas, Biological Warfare: Modern Offense and Defense (Boulder: Lynne Rienner Publishers, Inc., 2000),
p. 79-81.
2607 The reports that the group attempted to obtain an Ebola strain in Zaire in October 1992 were not born out by the current
authoritative study on Aum Shinrikyo by Danzig et al. that enjoyed unprecedented access to imprisoned top members of the
group.
Richard Danzig, Marc Sageman, Terrance Leighton, Lloyd Hough, Hidemi Yuki, Rui Kotani, Zachary M. Hosford, “Aum
Shinrikyo: Insights into how terrorists develop biological and chemical weapons, second edition,” Center for a New
American Security, December 2012
2608 For leader statements supporting the passive interest in Ebola, see:
D. W. Brackett, Holy Terror: Armageddon in Tokyo (New York: Weatherhill, 1996), p. 102;
Amy E. Smithson, “Rethinking the Lessons of Tokyo,” in Ataxia: The Chemical and Biological Terrorism Threat And The
US Response, ed. Amy E. Smithson, Stimson Report 35, October 9, 2000, p. 74, 74fn.12. <http://www.stimson.org/books-
reports/ataxia-the-chemical-and-biological-terrorism-threat-and-the-us-response/>
2609 Endo completed graduate courses, but dropped out before finishing his graduate degree, hence the use of the term “graduate-
trained.”
Amy E. Smithson, “Rethinking the Lessons of Tokyo,” p. 75;
edition,” p. 13, 23.
2610 Amy E. Smithson, “Rethinking the Lessons of Tokyo,” p. 75fn.15.
2611 Ibid.
2612 For instance, Fumihiro Joyu tried to recruit Russian chemical engineers with sarin nerve agent production experience.
edition,” p. 52.
2613 Ibid.

insider with sympathies for the cult provided him access to a non-pathogenic vaccine strain.2614 The
individual who allegedly did so was never found.2615 Independent interviews with two core Aum
Shinrikyo BW program members confirmed that they knew the strain obtained was non-pathogenic.2616
Endo had promised them that he could use “genetic engineering” to modify the strain to be pathogenic.2617
Richard Danzig et al., in their seminal study of the group’s WMD efforts, noted that Endo could have
attempted to exploit a known viable method to make a non-pathogenic strain pathogenic.2618 However, the
evidence that this method was attempted remains speculative. In any case, Endo proclaimed success, and
the group attempted several ineffectual attacks with B. anthracis.2619 Danzig et al. conclude by stating, “to
this day we nor the leaders of Aum Shinrikyo know whether Endo possessed a fully virulent strain of B.
anthracis and was unable to conserve it, or whether he conserved it but could not amplify it, or whether
he never achieved it at all.”2620
The B. anthracis production lines were crude. The first production method tried was a liquid line, set up
in 1992 at a cult property in Kameido, Tokyo.2621 The cult apparently relied on 200-liter drums to act as
fermenters, with 10 drums used for a production run.2622 No attempt was made at separate the pathogenic
culture from the growth media through liquid purification; the resultant slurry was used directly.2623 In
1993, following failed attacks with the liquid mixture, the group attempted to dry the product and
disseminate the resultant powder.2624 As before, no attempt was made to separate the growth media from
the pathogens.2625
Regarding delivery system capabilities, Aum Shinrikyo maintained a vehicle with a mounted spray-dryer
system for the biological and toxin programs, but the spray system was highly defective.2626 It was
employed once unsuccessfully with the group’s non-pathogenic B. anthracis.2627 The spray system was
manufactured by the cult themselves, because the group’s leader did not want to wait the two months
needed to order and receive a sprayer from a European firm.2628
16.10.4 Rajneesh Cult
The Rajneesh Cult was started by an individual calling himself Bhagwan Shree Rajneesh in India in the
1960s.2629 By 1981, the cult had moved to Wasco County, Oregon (U.S.A.).2630 By spring 1984, the cult
2614 No further details are known.

edition,” p. 23.
2615 Ibid.
2616 Ibid.
2617 Ibid.
2618 Ibid.
2619 Ibid.
2620 Ibid.
2621 Ibid.
2622 Ibid.
2623 Ibid.
2624 Ibid.
2625 Ibid.
2626 Ibid.
2627 Two failed attempts with botulinum toxin had previously been launched using the same platform. Ibid.
2628 Ibid.
2629 W. Seth Carus, “Rajneeshees,” Encyclopedia of Bioterrorism Defense, 2nd Edition, eds. Rebecca Katz, Raymond A.
2630 Ibid.

faced significant legal troubles, including land conflicts and immigration law violation investigations.2631
A highly influential member, Ma Anand Sheela, decided to devise a plan whose goal was the replacement
of two commissioners with cult members in the November 1984 election.2632 Ma Anand Puja, a senior
cult member close to Sheela and a registered nurse, was to organize a biological weapons attack to sicken
locals and thereby prevent them from voting.2633 They hoped that doing so would enable the cult’s
candidates to win the election despite their unpopularity in the local community.
A major “trial run” was carried out in September 1984, when S. typhimurium was used to contaminate at
least 10 restaurant salad bars in The Dalles, Oregon.2634 No individuals died, but at least 751 people fell ill
as a result of the attack.2635,2636 When the cult realized that they had no chance of winning the local
elections, they abandoned the planned November attack.2637 The September attack was misattributed by
health agencies as caused by unsanitary practices by restaurant workers for over a year.2638 The incident
only came to light because of a major rift between Bahgwan Rajneesh and Seela and Puja; the two women
fled the U.S. camp, and Bahgwan Rajneesh retaliated by publicizing their actions.2639
At least one biological attack was carried out before the major September attack.2640 On August 29, 1984,
two Wasco County commissioners were given water deliberately tainted with S. typhimurium, and both
fell ill.2641 Reports, based on admissions made by Rajneesh members, allege that other cult attacks took
place prior to August 1984, namely: one attack against the Wasco County Courthouse, attacks against
schools, nursing homes, and political gatherings, one attack against The Dalles’ water supply, and one
attack against a supermarket.2642,2643 These incidents are unconfirmed, since these alleged attacks
apparently did not cause illnesses or were not carried out by disobeying members, and since members
may have had a desire to exaggerate Puja’s wrongdoings given the aforementioned internal conflict and a
general dislike for Puja.2644,2645
Roughly fourteen individuals were associated with the cult’s biological weapons program: three or four
individuals were directly involved in culturing the Salmonella-causing pathogen for use in the September
attack, while seven or eight appear to have spread the biological agent (both teams had some overlap).2646
2631 Ibid.
2632 Ibid.
2633 Ibid.
2634 Ibid.
2635 Thomas J, et al. (1997) “A Large Community Outbreak of Salmonellosis Caused by Intentional Contamination of Restaurant
Salad Bars,” Journal of the American Medical Association 278, no. 5: 389-395,
http://www.cdc.gov/phlp/docs/forensic_epidemiology/Additional%20Materials/Articles/Torok%20et%20al.pdf.
Jonathan Tucker Cambridge: The MIT Press.
2637 W. Seth Carus, “Rajneeshees,” p. 534-535.
2638 Ibid.
2639 Ibid.
2640 Ibid.
2641 Ibid.
2642 Ibid.
Center for Counterproliferation Research, National Defense University, February.
2644 Ibid.
2646 Ibid.

The cult used common microbiological techniques to produce the desired quantities of agent: the
pathogen was grown on agar in Petri dishes and incubated.2647 Puja ordered the S. typhimurium sometime
between October 1, 1983 and February 29, 1984 from a medical supply company.2648 Puja had reportedly
also ordered cultures of S. typhi, Salmonella paratyphi, Francisella tularensis, and other undisclosed
pathogens from the American Type Culture Collection, using her status as a nurse at the Rashneeshee’s
state-licensed medical laboratory.2649 The decision to use S. typhimurium rather than these or other agents
was presumably made based on the desire to keep the attack covert, the desire to make people sick but not
kill them, and the relative ease of culturing the organism, although no accounts of Puja’s final decision
exists.2650
16.10.5 RISE
R.I.S.E. was a small domestic eco-radical group.2651 The group’s founders, the college students Stephen J.
Pera and Allen C. Schwandner, were arrested on January 18, 1972.2652 They had formed a group they
called R.I.S.E., and reportedly planned to contaminate Chicago’s municipal water system with Salmonella
typhi (causative agent of typhoid fever).2653 Precisely what the acronym R.I.S.E. stood for is not
known.2654 Two new recruits turned on the two founders and reported the plot to the police.2655 The
following account of the group’s activities come from publications by W. Seth Carus, who remains the
only researcher to have extensively studied this case to date.
16.10.5.1 Motivation and intent to use
Pera was an adopted child with a troubled childhood who repeatedly did not get along with others.2656 For
instance, he was asked to leave a microbiology program sponsored by the International Foundation for
Microbiology due to conflicts with other students.2657 Pera and Schwandner believed that mankind was
destroying the planet, and that the only way to prevent this was to wipe out the human race, except for a
chosen small group of like-minded individuals.2658 Despite reports to the contrary, the group had no neo-
Nazi or racist tendencies.2659 Although Pera and Schwandner eventually fled to Cuba, they did not appear
to have a particular affinity for communist countries; part of their planning involved striking the Soviet
Union and China, because they feared that these countries would capitalize on the group’s planned
destruction of the Western powers.2660 The group’s knowledge of eco-radical theory was rather primitive,
suggesting that R.I.S.E. was mostly a byproduct of their inability to adapt to society.2661
2648 Ibid.
2649 Ibid.
2650 Ibid.
Zilinskas Hoboken: John Wiley & Sons.
2652 Ibid.
2653 Ibid.
2654 Ibid.
2655 Ibid.
2656 Ibid.
2657 Ibid.
2658 Ibid.
2659 Ibid.
The R.I.S.E. case appears to be the exact same incident as one reported in passing in some secondary sources involving a
supposed far-right group called the “Order of the Rising Sun.” The plot details and timeline are identical. R.I.S.E. was not a
far-right group, but as Seth Carus has noted is sometimes flagged as such; there must have been some initial media
confusion, and this is perhaps what led to the “Order of the Rising Sun” story.
2660 Ibid.
2661 Ibid.

The group had obtained a range of pathogens, and considered aerosol and food contamination as
alternative dissemination pathways to the planned water supply attack.2662 Pera obtained S. typhi and N.
meningitides from a hospital microbiology laboratory where he volunteered.2663 C. diphtheria and S.
sonnei were also obtained by the group through some unknown means.2664 The group used this laboratory
to culture pathogens. In December 1971, Pera was kicked out of the lab for having tried to acquire
controlled substances illegally, and the hospital authorities destroyed his samples.2665 The group relocated
its activities to Mayfair College laboratories until the police arrested the two leaders.2666
In W. Seth Carus’ judgement, “although R.I.S.E. appears to have been motivated to conduct a mass-
casualty attack with biological weapons, it lacked the scientific and technical expertise to carry it out.”2667
The water supply contamination scheme would have failed had it been carried out, and although the group
talked about aerosol dissemination, the members had no relevant knowledge or experience.2668
Schwandner had no technical expertise. He enrolled at Mayfair College to study the humanities, but
rapidly stopped attending any of his classes.2669 Pera was largely self-taught in microbiology, with only
some low-level work experience to complement limited and incomplete coursework from Mayfair
College.2670 His cultures were found to have contained several organisms, demonstrating that he lacked
the skill necessary to prevent culture contamination.2671 Pera was the only member with any scientific
experience.2672
16.11 Other Terrorist/Extremist Groups Linked in Some Fashion to Biological Weapons
As discussed in Sec. 16.10, only five terrorist groups have sought a biological weapons
capability intended for mass casualty attacks. Another 14 groups have been linked in some fashion with
biological terrorism.2673 Four of these groups made apparently-empty threats. The count includes animal
rights extremist groups (two such groups are included). The count however excludes any groups involved
with toxin-only acts or threats. Very few information is available in open sources on several of the cases
below. The cases (numbered) are as follows:
2662 Ibid.
2663 Ibid.
2664 The group was originally suspected of having Botulinum toxin, but “subsequent tests […] indicated that the two did not
have any botulinum toxin.” W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p.
102.
2665 Ibid.
2666 Ibid.
2667 Ibid.
2668 Ibid.
2669 Ibid.
2670 The number of courses Pera signed up for and actually completed is unclear. Pera did not finish at least one course, and
appears to have signed on for at most two other courses.
Zilinskas (Hoboken: John Wiley & Sons.
2672 Ibid.
2673 In part based on W. Seth Carus’ “Appendix A: List of Cases:”
W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, Working Paper (Washington:
Center for Counterproliferation Research, National Defense University, February 2001 Revision), p. 179-198.
Note that the “World Islamic Front for Fighting Jews and Christians,” noted in one case, is an Al Qaeda alias.
U.S. Department of State, Bureau of Counterterrorism, “Country Reports on Terrorism 2013 - Chapter 6. Foreign Terrorist
Organizations,” April 2014, <http://www.state.gov/j/ct/rls/crt/2013/224829.htm>.

1- The eco-radical group “Dark Harvest” took B. anthracis-contaminated soil from Gruinard Island (a
then-contaminated British military World War II site used to test B. anthracis bombs) and spread it on the
grounds of Porton Down in 1981.2674 The soil did contain B. anthracis, although no harm resulted from
the act.2675 No further acts were attributed to this group, and it is presumed defunct.
2,3,4- Three ethno-nationalist groups have reportedly used biological agents to enhance the effectiveness
of conventional explosive attacks. Government forces have claimed that the FARC (Colombia), ELN
(Colombia), and NPA (Philippines) groups spiked explosive devices with feces to cause sepsis, in what
appears to be a modern take on the Viet Cong punji stick technique.2676,2677,2678 NPA has denied these
claims.2679 All three groups remain active and are designated as Foreign Terrorist Organizations by the
U.S. Department of State.
5- A Palestinian group (unknown) was reportedly caught in a counterfeiting scheme whereby expired
eggs contaminated with salmonella were stamped with counterfeit stamps indicating their acceptability to
be eaten, and sold.2680 Israeli news reporting on the group’s capture in May 2000 implied that this was
deliberately done to sicken Israelis.
6- The German-based, now-defunct, Red Army Faction (RAF) reportedly maintained a botulinum toxin
laboratory in Paris, France until it was uncovered in October 1980.2681 However, a recent review of this
case has cast doubt on parts of the underlying story, and German authorities apparently remain convinced
that “no evidence whatsoever [exists] that members of the `RAF’ had planned or prepared an attack using
biological agents.”2682,2683
7,8- Two animal rights radical groups have used the threat of HIV contamination to heighten fear. A
spokesman for the Animal Liberation Front (ALF) claimed in 1993 that bombs planted in the U.K. by
members of the collective had been purposefully tainted with HIV, but authorities dismissed this
account.2684 Similarly, the “Justice Department” mailed razors to fur retailers in Canada in 1996 which
they claimed were covered with HIV-infected blood, although whether they had really done so is not
known.2685
9- The “Counter Holocaust Lobbyists of Hillel” sent agar and B. cereus in a petri dish apparently labelled
2674 Porton Down was Britain’s main biodefense and chemical warfare defense establishment, and previously the center running
Britain’s biological weapons program. W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents
Since 1900, p. 58.
2675 Ibid.
2676 Pablo Esteban Parra Gallego, “IEDs: A Major Threat for a Struggling Society,” The Journal of ERW and Mine Action 13,
no. 3 (Winter 2009), <http://www.jmu.edu/cisr/journal/13.3/specialreport/gallego/gallego.shtml>;
5/articles/035c.htm
http://mindanaoexaminer.com/philippine-army-finds-human-feces-snake-venom-in-wounded-soldiers-wounds/.
2679 Ibid.
2680 Jason Pate, Gavin Cameron, “Covert Biological Weapons Attacks against Agricultural Targets: Assessing the Impact against
U.S. Agriculture,” BCSIA Discussion Paper 2001-9, ESDP Discussion Paper ESDP-2001-05, John F. Kennedy School of
Government, Harvard University, August 2001, p.8,
http://belfercenter.ksg.harvard.edu/files/covert_biological_weapons_attacks_against_agricultural_targets.pdf.
2681 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 156-157.
2682 Ibid
2683 The review in question is:
Terence Taylor, Tim Trevan, “The Red Army Faction (1980),” Toxic Terror: Assessing Terrorist Use of Chemical and
Biological Weapons, ed. Jonathan B. Tucker (Cambridge: MIT Press, 2000), p. 107-113.
2685 Ibid.

“anthracs” [SIC] and “Yersinia” to a Jewish organization in Washington in 1997.2686,2687 The package
contained a hate latter that further misrepresented the petri dish as containing a “chemical warfare”
agent.2688 Whether this was an anthrax hoax, or whether the group thought the package contained B.
anthracis, is unknown. No further acts were attributed to this group, and it is presumed defunct.
10- The Chinese government alleges that Emeti Yakuf, an alleged terrorist connected to the East
Turkistan Islamic Movement, threatened to use biological and chemical weapons to disrupt the 2008
Olympics held in China, and that he trained group members on making poisons.2689 This individual was
reportedly killed in a 2012 U.S. drone strike in Pakistan.2690
11 to 14- Another four groups have reportedly threatened to use a biological agent, but did not specify
what type of agent, and are not known to have possessed biological agents. These were: Chechen
separatists (in general), the “Republic of Texas” group, the Al-Aqsa Martyrs Brigade, and the “Indian
Mujahedeen (Assam).”2691,2692,2693
16.12 The Islamic State of Iraq and the Levant (ISIL)
ISIL (also called ISIS or Daesh) is a Sunni violent Islamist group currently in control of territory in Syria,
Iraq, and Libya. It is fighting against the Iraqi government’s mostly-Shia forces, and is a major player in
the Syrian civil war.2694 It seeks to establish and expand its own state: a caliphate with its leader, Abu
Bakr al-Baghdadi, as caliph.2695,2696 Public estimates of the group’s fighting strength vary tremendously,
from a low of 20,000 to a high of 200,000 fighters.2697,2698,2699 As explained in a report generated by the
United Nations’ Analytical Support and Sanctions Monitoring Team for the United Nations Security
2686 Ibid.
2687 The B’nai B’rith International Jewish Monthly, Volume 111, (1996), p. 67, https://books.google.com/books?id=V--
2688 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 111;
Matthew Dorf, “Questions lingering after bizarre mailing to B’nai B’rith,” J Weekly, May 2, 1997,
http://www.jweekly.com/article/full/5673/questions-lingering-after-bizarre-mailing-to-b-nai-b-rith/.
2689 “Èastern Turkistan’ terrorists identified,” China Daily, October 21, 2008, http://www.chinadaily.com.cn/china/2008-
10/21/content_7126503.htm.
2690 Declan Walsh, Eric Schmitt, “Militant Leader Believed Dead in Pakistan Drone Strike,” The New York Times, August 24,
2012, http://www.nytimes.com/2012/08/25/world/asia/us-drone-strikes-kill-18-in-pakistan.html?_r=1.
2691 W. Seth Carus, Bioterrorism and Biocrimes: The Illicit Use of Biological Agents Since 1900, p. 107-109, 186.
Gary Ackerman, Jeremy Tamsett (Boca Raton: CRC Press, 2009), p. 143
http://www.nti.org/gsn/article/extremists-warn-of-biological-strike-in-india/.
2694 “Islamic State- The Pushback,” The Economist, March 21, 2015, <http://www.economist.com/news/briefing/21646752-
sustaining-caliphate-turns-out-be-much-harder-declaring-one-islamic-state-not#correction>.
2695 United Nations Security Council, “Letter dated 13 November 2014 from the Chair of the Security Council Committee
pursuant to resolutions 1267 (1999) and 1989 (2011) concerning Al-Qaida and associated individuals and entities addressed
to the President of the Security Council,” S/2014/815, November 14, 2014, p.6-7, para. 7, 12,
http://www.securitycouncilreport.org/atf/cf/%7B65BFCF9B-6D27-4E9C-8CD3-CF6E4FF96FF9%7D/s_2014_815.pdf.
2696
‫ ب يروت‬- ‫[ ب ف أ‬Beirut – AFP], "‫“[ ال ب غدادي وي باي ع "إ س الم ية خ الف ة" ق يام ي ع لن "داعش‬ISIS” declares the “Islamic caliphate,”
pledges allegiance to al-Baghdadi], Alhayat, June 29, 2014, http://www.alhayat.com/Articles/3292478
to the President of the Security Council,” p. 8, para. 14.
2698 “Islamic State formations comprise up to 70,000 gunmen- Chief of Russia’s General Staff,” TASS, December 10, 2014,
http://tass.ru/en/world/766237.
2699 Patrick Cockburn, “War with Isis: Islamic militants have army of 200,000, claims senior Kurdish leader,” The Independent,
November 16, 2014, http://www.independent.co.uk/news/world/middle-east/war-with-isis-islamic-militants-have-army-of-

200000-claims-kurdish-leader-9863418.html.

Council, part of the uncertainty stems from the lack of clarity as to “whether all those fighting with ISIL
[…] have actually pledged loyalty to the group, or are in allied militia groups, or are opportunistically
aligning with ISIL, or have been forced to fight.”2700 Compounding these issues, all available estimates
are now dated, having been issued at the end of 2014. Estimates of the number of individuals living in
areas under ISIL control are also uncertain: cited figures include “about 8 million” and “10 million
people.”2701,2702
The group’s leadership is dominantly Iraqi, given that ISIL evolved from Abu Musab al-Zarqawi’s Al
Qaeda in Iraq. Zarqawi had serious strategic disagreements with Osama bin Laden’s Al Qaeda central,
starting in February 2004, over the former’s desire to heavily target Iraq’s Shia population and thereby
incite sectarian violence.2703 Al-Zarqawi was subsequently killed in a U.S. airstrike in June 2006.2704 Abu
Bakr al-Baghdadi took control of the group in 2010, and the group changed names in 2013 and
2014.2705,2706 Reconciliation efforts between ISIL and Al Qaeda central failed, and the latter formally
disassociated itself from ISIL in February 2014.2707
The aforementioned U.N. report noted that ISIL is “particularly well-armed given its access to extensive
supplies of heavy weapons seized from the Government of Iraq,” and “has fighters with experience in
conventional warfare who are well-versed on a range of weapons systems, including the use of tanks and
artillery.”2708 Group propaganda has displayed numerous heavy weapons in use, including anti-tank
missile systems.2709,2710,2711,2712
to the President of the Security Council,” p. 8, para. 14.
2701 The 10 million estimate was given by Peter Maurer, president of the International Committee of the Red Cross, in March
2015. Stephanie Nebehay, “Islamic State-controlled parts of Syria, Iraq largely out of reach: Red Cross,” Reuters, March 13,
2015, <http://www.reuters.com/article/2015/03/13/us-mideast-crisis-syria-icrc-idUSKBN0M921N20150313>;
2702 The “about 8 million” estimate is given in: “Islamic State- The Pushback,” The Economist.
For an analysis of issues with generating these figures, see: Frank R. Gunter, “The ISIL Invasion of Iraq: Economic Winners
and Losers,” Foreign Policy Research Institute, July 2014, <http://www.fpri.org/articles/2014/07/isil-invasion-iraq-
economic-winners-and-losers>.
2703 Emily Hunt, “Zarqawi’s `Total War’ on Iraqi Shiites Exposes a Divided among Sunni Jihadists,” The Washington Institute,
PolicyWatch #1049, November 15, 2005, <http://www.washingtoninstitute.org/policy-analysis/view/zarqawis-total-war-on-
iraqi-shiites-exposes-a-divide-among-sunni-jihadists>.
2704 Ellen Knickmeyer, Jonathan Finer, “Insurgent Leader Al-Zarqawi Killed in Iraq,” The Washington Post, June 8, 2006, p.1-2,
<http://www.washingtonpost.com/wp-dyn/content/article/2006/06/08/AR2006060800114.html>.
2705 Initially to the “Islamic State in Iraq and al-Sham,” in April 2013; subsequently to the “Islamic State,” in June 2014. Jessica
D. Lewis, “Al-Qaeda in Iraq Resurgent: The Breaking The Walls Campaign, Part I,” Institute for the Study of War,
September 2013, p. 9, <http://www.understandingwar.org/sites/default/files/AQI-Resurgent-10Sept_0.pdf>;
to the President of the Security Council,” p. 7 para. 11, 12.
2707 Liz Sly, “Al-Qaeda disavows any ties with radical Islamist ISIS group in Syria, Iraq,” The Washington Post, February 3,
2014, https://www.washingtonpost.com/world/middle_east/al-qaeda-disavows-any-ties-with-radical-islamist-isis-group-in-
syria-iraq/2014/02/03/2c9afc3a-8cef-11e3-98ab-fe5228217bd1_story.html.
to the President of the Security Council,” p. 14 para. 37.
2709 “Open Syria @OpenSyria” [Twitter handle / Pseudonym], “IS TOW use geo-located ~2.3km NE of Palmyra, documented in
“The Raid of Abu Malik A-Tamimi […]” Twitter, June 8, 2015, https://twitter.com/OpenSyria/status/607888537115987968
2710 “Open Syria @OpenSyria” [Twitter handle / Pseudonym], “#IS Kornet ATGM deployed at the #Hasakah prison, reported
tank kill […],” Twitter, June 2, 2015, https://twitter.com/OpenSyria/status/605641058446278656;
2711 “Open Syria @OpenSyria” [Twitter handle / Pseudonym], “9M14M Malyutka (perhaps Iranian I-Raad) ATGM among #IS
spoils in new Wilayat #Hama release […],” Twitter, June 1, 2015,
<https://twitter.com/OpenSyria/status/605408366068760576>;
2712 “The Islamic State’s spring offensive: al-Sukhna,” Oryx Blog, May 23, 2015, http://spioenkop.blogspot.com/2015/05/the-
islamic-states-spring-offensive-al.html.

The group has co-opted, recruited, or coerced numerous specialists. In one notable case of co-optation,
Bashar al-Assad regime’s technical experts maintaining the Euphrates dam near Raqqa have remained on
site on the government’s payroll, allegedly in exchange for the continued operation of the ISIL-controlled
dam.2713,2714,2715 ISIL used engineers to operate the oil refineries it initially controlled, although these
operations have been heavily disrupted by coalition airstrikes against ISIL-controlled refineries and
related transport convoys.2716,2717 A February 2015 report by the U.S. Financial Action Task Force noted
that ISIL controlled “numerous oil fields from which it continue[d] to extract oil for its own use [and] its
own refining,” even though the group lacked the “resources and technical capacities” to fully exploit these
resources.2718 The group has also demonstrated its chemical engineering capabilities through the
smuggling of chemicals such as phosphate. The possibility of such an event was raised by the FATF
report authors, who remarked that the Akashat Phosphate Mine and the Al-Qaim (sulfuric acid and
phosphoric acid) Manufacturing Plant had reportedly fallen under ISIL control.2719 By June 2015, an
anonymous analyst had released to the public satellite imagery showing the Al-Qaim facility pre- and
post-ISIL control, demonstrating that hundreds of tons of phosphate had been drained.2720 ISIL members
have generated propaganda specifically calling for specialists; one such appeal read: “we need engineers,
we need doctors, we need professionals. Every person can contribute something.”2721 ISIL members
recruited Western medical professionals, in part through propaganda portraying “really good medical
service” in occupied areas and calling upon medical students to join in building a new society.2722
The possibility that the group could harness its resources and human technical base to develop and
employ unconventional weapons has been raised in public reports numerous times.2723 Several claims that
ISIL is employing readily available dangerous chemicals as chemical weapons in Syria have been
published in the media.2724 Strong open source information supporting these claims emerged from a series
of mortar shell attacks that occurred in June and July 2015. A 120-millimeter mortar shell modified to
disseminate a chemical agent, “most probably chlorine,” was found in Kurdish positions.2725 It was
2713
Yezid Sayigh, “The War Over Syria’s Gas Fields,” Carnegie Endowment for International Peace, June 8, 2015,
http://carnegieendowment.org/syriaincrisis/?fa=60316;
2714 Danya Chudacoff, “`Water war’ threatens Syria lifeline,” Al Jazeera, July 7, 2014,
<http://www.aljazeera.com/news/middleeast/2014/07/water-war-syria-euphrates-2014757640320663.html>;
2715 Jan Ali, “Euphrates Dam… another victim of Syrian war,” ARA News, December 6, 2014,
<http://aranews.net/2014/12/euphrates-dam-another-victim-syrian-war/>.
2716 Fazel Hawramy, Shalaw Mohammed, Luke Harding, “Inside Islamic State’s oil empire: how captured oilfields fuel Isis
insurgency,” The Guardian, November 19, 2014, <http://www.theguardian.com/world/2014/nov/19/-sp-islamic-state-oil-
empire-iraq-isis>;
2717 Financial Action Task Force (FATF), “Financing of the Terrorist Organisation Islamic State in Iraq and the Levant (ISIL),”
February 2015, p. 13, <http://www.fatf-gafi.org/media/fatf/documents/reports/Financing-of-the-terrorist-organisation-
ISIL.pdf>.
2718 Financial Action Task Force (FATF), “Financing of the Terrorist Organisation Islamic State in Iraq and the Levant (ISIL),”
p. 13.
2719 Ibid, p. 16.
2720 “Not a spy @finriswolf” [Twitter handle / Pseudonym], “#Iraq :Unpublished Imagery: #ISIS has removed hundreds of tons
of phosphate from this facility […],” Twitter, June 7, 2015, <ttps://twitter.com/finriswolf/status/607466224927186944>.
2721 Liezel Hill, Scott Deveau, Gerrit De Vynck, “Canadians from Calgary to Timmins heed ISIL’s tweets,” Bloomberg, October
23, 2014, <http://www.bloomberg.com/news/articles/2014-10-23/canadians-from-calgary-to-timmins-heed-islamic-state>.
2722 Katrin Bennhold, “Young Medics Were Lured by Briton to Join ISIS,” The New York Times, July 17, 2015,
<http://www.nytimes.com/2015/07/18/world/europe/young-medics-were-lured-by-briton-to-join-
isis.html?rref=world/middleeast>.
2723 See for example: David Albright, Serena Kelleher-Vergantini, Sarah Burkhard, “Syria’s Unresolved Nuclear Issues
Reemerge in Wake of ISIL Advance and Ongoing Civil War,” Institute for Science and International Security – Imagery
Brief, June 30, 2015, p. 1-7, <http://isis-online.org/uploads/isis-reports/documents/Syria_June_30_2015_Final.pdf>.
2724 C. J. Chivers, “ISIS Has Fired Chemical Mortar Shells, Evidence Indicates,” New York Times, July 17, 2015,
<http://www.nytimes.com/2015/07/18/world/middleeast/islamic-state-isis-chemical-weapons-iraq-syria.html?smid=tw-
share>.
2725 Ibid.

analyzed by a military ordnance expert working for Sahan Research in partnership with Conflict
Armament Research.2726 Another similar mortar shell, analyzed by an expert from Conflict Armament
Research, apparently contained phosphine and had also been fired against Kurdish forces.2727 U.S.
intelligence agencies reportedly have concluded that ISIL used mustard agent in subsequent attacks in
Syria and Iraq.2728,2729
Despite these alarming trends, no credible, open source evidence exists confirming whether ISIL is
seeking a biological weapons capability. A journalistic piece in Foreign Policy described the contents of
an alleged ISIL member’s laptop hard drive, obtained by journalists for Foreign Policy, was found to
contain over 35,000 files dedicated to Jihad, a few of which discussed BW.2730 However, whether the
alleged owner intended to act on the BW files is not known. Moreover, the contents as described have no
grounding in technicality, and read like extracts from extremist “weapons cookbook” literature available
on the World Wide Web. For instance, the snippet of text from the file made public was: “Use small
grenades with the virus, and throw them in closed areas like metros, soccer stadiums, or entertainment
centers. Best to do it next to the air-conditioning. It also can be used during suicide operations.”2731 The
text mischaracterizes Y. pestis as a virus, and seemingly provides no instructions on how to create a
“small grenade” that would disseminate the agent successfully.
16.13 Biosafety and Biosecurity at U.S. Research Laboratories
The requirements that govern U.S. laboratory operations primarily derive from four sources: statutes,
regulations, guidance, and contracts. (Appendix V: Sec 16.11 lists all governing documents that are
relevant to the biosecurity risk assessment). Many security requirements are regulatory while biosafety
requirements are either contractual, associated with inspections of regulatory programs, or voluntary.
Furthermore, standards in guidance documents often are considered de facto requirements, especially
when needed for facility certification, regulatory compliance, liability protection, and compliance with
funding contracts or terms and conditions of grant awards.
To build a realistic picture of defense measures, both requirements and practice are considered.
Recognizing that institutions may be subject to state, local, and tribal requirements and institutional
policies all of which vary, the evaluation of defensive measures is based solely on federal governing
instruments and their implementation. Indications of practice may be gleaned from guidance documents,
peer-reviewed literature, other publically available sources, and interviews of officials representing
institutions conducting relevant research. However, no overarching documentation of industry best
practices exists, even though the desire to create such a mechanism has been voiced.
Figure 16.3 highlights required and voluntary measures at for non-select agent high containment
laboratories, select agent laboratories, and Tier 1 select agent laboratories. Some of the voluntary
measures, such as institutional threat assessment teams and surveillance of animal facilities, apply to non-
high containment laboratories.
2726 Ibid.
2727 Phosphine, chemical formula PH3, is used as a fumigant, but is toxic if inhaled. Ibid; also see: Sajila Saseendran, “Ministry
mulls banning `killer’ pesticide,” Khaleej Times, September 2, 2014,
<http://www.khaleejtimes.com/article/20140901/ARTICLE/309019899/1002>.
2728 Nabih Bulos, “Islamic State confirmed to have used mustard gas against Kurds in Syria,” The Telegraph, August 15, 2015,
<http://www.telegraph.co.uk/news/worldnews/middleeast/syria/11805235/Islamic-State-confirmed-to-have-used-mustard-
gas-against-Kurds-in-Syria.html>;
2729 Paul Blake, “US official: ÌS making and using chemical weapons in Iraq and Syria’,” BBC News, September 11, 2015,
<http://www.bbc.com/news/world-us-canada-34211838>.
<http://foreignpolicy.com/2014/08/28/found-the-islamic-states-terror-laptop-of-doom/>.
2731 Ibid.

Figure 16.3. Federal Select Agent and Toxin Program requirements are in addition to general infectious agent
biosafety and biosecurity requirements. These requirements represent the minimum security standards for
institutions. Many institutions implement additional safety and security measures, some of which are
included in the figure.
16.13.1 Biosafety Levels, Select Agents, and Risk Assessments
The laboratory operating environment can be broadly characterized using a tiered, agent-specific, and
experiment-specific framework. This framework draws on two classification systems: 1) Biosafety Levels
(BSLs) specifying containment, access, and security measures; and 2) the Federal Select Agent Program
that requires special safety and security precautions for designated agents.2732,2733 Under this framework,
an institution carries out a risk assessment before all planned experiments with pathogens to identify the
safety and, if applicable, the security risk of the experiment. The Federal Select Agent Program (FSAP)
requires that institutions and individuals seeking access to agents on the Biological Select Agents and
Toxins (BSAT) list be approved by the FSAP and under conditions specified by the FSAP. The biosafety
risk assessment helps the institution implement the appropriate measures necessary to mitigate risk and
comply with statutes and regulations. The Biosafety Levels, the FSAP, and the risk assessment process
are described in turn below.
2732 Federal Select Agent and Toxin Program http://selectagents.gov.

2733 Biosafety levels were originally established in the 1970s. Current updates can be found in the latest version of the BMBL.
Nancy Connell, “Biological Agents in the Laboratory- The Regulatory Issues,” Public Interest Report [Federation of
American Scientists] 64, no. 3 (Fall 2011): p. 13, http://fas.org/pubs/pir/2011fall/2011FALL-PIR-lowres.pdf.

16.13.1.1 Biosafety Levels
Biosafety Levels described in the BMBL are a means to categorize laboratory containment capabilities
based on facility specifications, safety equipment, and microbiological practices. BSLs range from lowest
(BSL-1) to highest (BSL-4) levels of containment.2734,2735 Laboratories that work with experimentally
infected animals require special measures at all levels compared to laboratories that do no such work. To
make the distinction clear, animal labs are categorized in a similar manner, using an Animal Biosafety
Level 1-4 scale that ranges from ABSL-1 (lowest containment) to ABSL-4 (highest containment).2736
USDA further established the BSL-3-Agriculture (-Ag), BSL-3 Enhanced, and ABSL-3 Enhanced levels
to describe special measures to reduce risk of environmental contamination when working with certain
livestock and plant pathogens.2737 A full description and comparison of Biosafety Levels can be found in
the BMBL.2738
The biosafety levels describe safety-specific measures primarily intended to prevent laboratory-acquired
infections and environmental release. However, certain measures also directly reduce security risks. For
instance, physical barriers and access controls serve a dual safety and security purpose. Figure 16.4 below
highlights some of the similarities and differences between the various biosafety levels on topics of
relevance to security. The table focuses on security-related measures and does not characterize the full set
of requirements for the biosafety levels. Selected biosafety measures are incorporated into broad security-
related categories: physical security, surveillance and monitoring, personnel training and reliability, and
emergency response.2739
2734 Ibid.
2735 U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories – Fifth Edition,
p. 24.
2736 Ibid.
2737 Ibid.
2738 Ibid.
2739 These are the subset of security categories presented in section 1.9 which appear in BSL recommendations.

Figure 16.4: Security-relevant requirements for BSL-1 through BSL-4, and ABSL-1 through ABSL-4, laboratories2740
2740 U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories – Fifth Edition, p.30-103, 343-350.

16.13.1.2 Biological Select Agents and Toxins
The list of Biological Select Agents and Toxins designate the pathogens and toxins2741 that pose a high
threat to human, animal, or plant health, and, hence, require special additional biosafety and biosecurity
measures.2742 The Antiterrorism and Effective Death Penalty Act of 1996 established the Select Agent
Program required restricted transfer of certain biological agents, now known as BSAT.2743 The current
Select Agents program was created through a significant expansion of this initial system in response to
the September 11 attacks and the “Amerithrax” attacks of 2001.2744,2745,2746
The BSAT list classifies pathogens depending on the disease host (human, animal, overlap human and
animal, and plant). HHS maintains a list of HHS Select Agents causing disease in humans including those
on the overlap list, the U.S. Department of Agriculture (USDA) maintains a list of Veterinary Service
Select Agents causing disease in animals including those on the overlap list, and the Animal and Plant
Health Inspection Service (APHIS) maintains a list of Plant Protection and Quarantine Select Agents.2747
Although the pathogens covered by each of these lists differ, the additional safety and security
requirements mandated in the relevant Parts of the U.S. Code of Federal Regulations for select agents are
functionally equivalent.2748,2749,2750,2751 That is, although the governing agency may change depending on
the pathogen considered, the security requirements for any BSAT are identical in practice. For this reason,
agents on any of these three lists are typically referred to as “Biological Select Agents and Toxins,”
without reference to a particular list.
The Select Agent Program further designates certain Select Agents as Tier 1 Select Agents.2752 These
agents “have the potential to pose a severe threat to public health and safety” (HHS) or “pose a severe
2741
Regulations involving toxins are not within the scope of this report, and references that apply solely to toxins are therefore
omitted.
2742 Nancy Connell, “Biological Agents in the Laboratory- The Regulatory Issues,” p. 14.
2743 These were initially called Listed Biological Agents, and were solely agents with “the potential to pose a severe threat to
public health and safety” as determined by the HHS Secretary.
Antiterrorism and Effective Death Penalty Act of 1996, Public Law 104-132, 104th Congress, Subtitle B--Biological
Weapons Restrictions, Sec. 511, http://www.gpo.gov/fdsys/pkg/PLAW-104publ132/html/PLAW-104publ132.htm.
2744 A short history can be found at the Select Agent program webpage: Centers for Disease Control and Prevention (CDC),
Animal and Plant Health Inspection Service (APHIS), “History,” http://www.selectagents.gov/history.html.
2745 USA PATRIOT Act of 2001, Public Law 107-56, 107th Congress, Title VIII—Strengthening the Criminal Laws Against
Terrorism, Sec. 817, http://www.gpo.gov/fdsys/pkg/PLAW-107publ56/html/PLAW-107publ56.htm.
2746 Public Health Security and Bioterrorism Preparedness and Response Act of 2002, Public Law 107-188, 107th Congress,
Subtitle D—Criminal Penalties Regarding Certain Biological Agents and Toxins, Sec. 231,
http://www.gpo.gov/fdsys/pkg/PLAW-107publ188/html/PLAW-107publ188.htm.
2747 Pathogens can be on both the HHS and USDA Select Agents lists, and are termed Overlap Select Agents.
2748 The American Biological Safety Association described the USDA select agents requirements as "essentially identical to the
requirements for select agents regulated by HHS." A review of the relevant regulations supports this statement.
American Biological Safety Association, “Re: Federal Register Docket CDC-2012-0010,” December 14, 2012,
http://www.absa.org/pdf/121214ABSACommentsCDC-2012-0010.pdf.
2749 U.S. Government Publishing Office, “Title 42: Public Health, §73.3 HHS select agents and toxins,” www.ecfr.gov/cgi-
bin/text-idx?SID=27b43dad6d0e40ba856cd39358931a6f&mc=true&node=se42.1.73_13&rgn=div8.
2750 U.S. Government Publishing Office, “Title 9: Animals and Animal Products, §121.3 VS select agents and toxins,”
idx?SID=e84486cd28bdb8517340f1c8b365ba9c&mc=true&node=se9.1.121_13&rgn=div8.
2751 U.S. Government Publishing Office, “Title 7: Agriculture, §331.3 PPQ select agents and toxins,” http://www.ecfr.gov/cgi-
bin/text-idx?SID=a2965b1fa4298b718b9259a19efe533f&mc=true&node=se7.5.331_13&rgn=div8.
2752 The risk-based tiering of the Select Agents lists was mandated through Executive Order 13546, Sec. 4. Executive Order
13546, “Optimizing the Security of Biological Select Agents and Toxins in the United States” (July 2010)
http://www.gpo.gov/fdsys/pkg/FR-2010-07-08/pdf/2010-16864.pdf.

threat to animal health or to animal products” (USDA).2753,2754 Tier 1 Select Agents have additional safety
and security requirements that go beyond what is required for non-Tier 1 Select Agents and Toxins.
Three pathogens have additional special safety and security requirements that go beyond the Tier 1
requirements: variola (major and minor) virus, rinderpest virus, and foot-and-mouth disease virus. These
pathogens are not within the scope of the current report. Therefore, the specific measures for these
pathogens are not summarized below, although their special security measures have been examined
during our review of the possible security landscape and in considering potential recommendations.
BSAT requirements are in addition to the general infectious agent biosafety measures (Figure 16.3).
Under the current requirements, MERS-CoV and low pathogenic avian influenza (LPAI) are not BSAT.
SARS-CoV, highly pathogenic avian influenza (HPAI), and recombinant 1918 influenza are Select
Agents. The CDC has recently proposed categorizing laboratory-modified H5N1 influenza virus strains as
Tier 1 Select Agents.2755
For all infectious agent research, the operating environment is primarily defined by the selection of a
biosafety level. For work with a Select Agent or Tier 1 Select Agent, additional safety and security
requirements will significantly affect operations.2756 The combination of biosafety level and Select Agent
status define a framework within which all security-related requirements and practices can be analyzed.
Most GoF research is conducted at various types of Level 3 containment (BSL-3, BSL-3 Enhanced,
ABSL-3, BSL-3-Ag, or ABSL-3 Enhanced). Some may occur under Level 2 containment with additional
respiratory protection. GoF research currently uses non-Select and Select Agents, though some
recombinant influenza strains may be reclassified as Tier 1 Select Agents in the near future.
16.13.1.3 The Risk Assessment Process
The risks of a given research plan are agent-specific and experiment-specific. Risk assessments are a key
part of experiment planning. Biosafety risk assessments (also known as biological risk assessments) are
required for all infectious agent research following BMBL practices, and also under OSHA and NIH
guidelines.2757,2758,2759 Currently, no federal regulations explicitly require biosecurity risk assessments for
research with non-Select Agents, although the BMBL provides advisory recommendations for principles
2753 U.S. Government Publishing Office, “Title 42: Public Health, §73.3 HHS select agents and toxins,” www.ecfr.gov/cgi-
bin/text-idx?SID=27b43dad6d0e40ba856cd39358931a6f&mc=true&node=se42.1.73_13&rgn=div8.
2754 U.S. Government Publishing Office, “Title 9: Animals and Animal Products, §121.3 VS select agents and toxins,”
idx?SID=e84486cd28bdb8517340f1c8b365ba9c&mc=true&node=se9.1.121_13&rgn=div8.
2755 Proposed regulation covers laboratory generated, mammalian, respiratory-transmissible influenza viruses containing the
hemagglutinin from the A/Goose/Guangdong/1/96 lineage. Federal Register Volume 80, Number 136, Pages 42079-42084
http://www.gpo.gov/fdsys/pkg/FR-2015-07-16/html/2015-17435.htm.
2756 Work with Select Agents and Toxins must still satisfy all regular biosafety requirements for infectious agent work.
p. 9;
2758 U.S. Code, Title 29, Chapter 15-Occupational Safety and Health, Section 654. http://www.gpo.gov/fdsys/pkg/USCODE-
2010-title29/html/USCODE-2010-title29-chap15-sec654.htm.
2759 National Institutes of Health, “NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules
(NIH Guidelines),” November 2013, http://osp.od.nih.gov/sites/default/files/NIH_Guidelines_0.pdf.

of laboratory biosecurity.2760,2761 Select Agent regulations require the development, implementation, and
regular review of a security plan, which is designed according to a site-specific risk assessment.2762
16.13.1.3.1 Biosafety risk assessments
The biosafety risk assessment process involves laboratory directors, institutional biosafety committees,
biological safety professionals, institutional review boards, animal care and use committees, and animal
facility directors.27632764 These risk assessors choose an appropriate biosafety level for an experiment by
considering the infectivity of the pathogen, the severity of the disease it causes, its transmissibility,
whether the pathogen is indigenous or exotic, and the nature of the work to be conducted.2765 The BMBL
recommends “careful judgment” during the risk assessment process; underestimating risk can be
dangerous, but overprescribing measures may add expense, make research more logistically difficult, and
lead to noncompliance.2766
Baseline biosafety level recommendations for many infectious agents are provided in the BMBL and
through CDC and WHO guidance.2767 Current BMBL and CDC guidance recommends that virus
propagation in cells or animals occur in Level 3 containment (Standard, Animal, Enhanced, or
Agricultural) for highly pathogenic avian influenza (HPAI), recombinant 1918 influenza, non-
contemporary H2N2, SARS-CoV, and MERS-CoV; Level 2 containment (Standard or Animal) is
recommended for low pathogenic avian influenza (LPAI) and currently circulating seasonal
influenza.27682769 The BMBL also strongly recommends a thorough risk assessment is conducted before
starting any experiments where pathogenic characteristics are deliberately enhanced, as specific guidance
is based on an infectious agent’s “capability to infect and cause disease.”2770
In addition to these broad risk factors, the BMBL highlights specific factors to consider during biosafety
risk assessments for influenza virus research. These factors are: replication in the respiratory tract in
animal models; clonal purity; phenotypic stability; gene constellations; and time since a similar strain was
circulating widely in nature. Although these factors do not directly apply to research with the other agents
2760 Additional review is required for funding of some work under Dual Use Research of Concern policy (DURC). However, all
GoF-relevant pathogens on the DURC list are also Select Agents.
2761 U.S. Department of Health & Human Services, “United States Government Policy for Oversight of Life Sciences Dual Use
Research of Concern,” March 2012, http://www.phe.gov/s3/dualuse/Documents/us-policy-durc-032812.pdf.
2762 Centers for Disease Control and Prevention (CDC) Division of Select Agents and Toxins, Animal and Plant Health
Inspection Service (APHIS) Agriculture Select Agent Program, “Security Guidance for Select Agent or Toxin Facilities: 7
CFR Part 331, 9 CFR Part 121, 42 CFR Part 73,” July 5, 2013,
http://www.selectagents.gov/resources/Security_Guidance_v3-English.pdf.
2763 U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories – Fifth Edition.
2764 This system was codified for civilian research in 1974, through CDC’s Classification of Etiologic Agents on the Basis of
Hazard. U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories – Fifth
Edition, p. 3.
2765 Ibid.
2766 Ibid.
2767 Ibid.
For example, CDC provided the following guidance on MERS: Centers for Disease Control and Prevention (CDC), “Middle
East Respiratory Syndrome (MERS),” June 18, 2015, http://www.cdc.gov/coronavirus/mers/guidelines-lab-biosafety.html.
2768 Whereas guidance covers preparations such as fixed samples and untreated diagnostic specimens, we focus here on cellular
and animal propagation of highest interest to the GoF research community. Certain experiments may require additional
safety measures (e.g. respiratory protection at BSL-2). U.S. Department of Health and Human Services, Biosafety in
Microbiological and Biomedical Laboratories – Fifth Edition, p. 211, 224.
2769 CDC, “Interim Laboratory Biosafety Guidelines for Handling and Processing Specimens Associated with Middle East
Respiratory Syndrome Coronavirus (MERS-CoV) – Version 2,” June 18, 2015,
http://www.cdc.gov/coronavirus/mers/guidelines-lab-biosafety.html.
p. 10.

of interest– including coronaviruses– they do provide insight into some of the factors generally under
consideration during biosafety risk assessments.
In addition, a laboratory manager may deem additional security measures necessary for work with non-
select agents that nevertheless pose a “high public health and agriculture concern,” or with commercially
valuable products such as vaccine candidates.2771
16.13.1.3.2 Biosecurity risk assessments
Although biosecurity risk assessments are not required for a majority of non-Select Agent research, they
may still be implemented. The BMBL recommends considering adversaries, threats, and scenarios whilst
developing written security plans, standard operating procedures, incident response plans, and employee
training protocols.2772 Other sources recommend considering physical security, personnel reliability,
material control and accountability, and information security.2773
Select Agent guidance notes that risk assessments are “the cornerstone of a good security plan.”2774 Risk
assessments should be performed by a team that includes the responsible official, biological safety
professionals, lead investigators, facility security and operations, federal partners, and local law
enforcement. This team should assess malicious actor threats, natural hazards, consequences, and
particular vulnerabilities, and develop a plan to mitigate identified risks.2775 Tier 1 Select Agents require
additional security measures, but the risk assessment process is the same.
The September 2014 institutional DURC oversight policy requires research institutions to conduct a risk
assessment of proposed research to determine whether it falls within the policy’s definition of “dual use
research of concern.”2776 This risk assessment is a three-step process: 1) determining whether the
proposed research involves one of the 15 listed agents; 2) evaluating whether the proposed research
involves one of seven categories of experiments; and 3) assessing the consequences of the research to
determine whether it qualifies as “dual use research of concern.” If proposed research is thought to have
dual use potential, the principal investigator and institution are required to identify appropriate risk
mitigation plans according to the responsibilities enumerated in the institutional DURC policy.2777
p. 104-105.
2772 Ibid.
2773 LouAnn C. Burnett, “Biosafety Practices Associated with Potential Agents of Biocrime and Biowarfare,” Current Protocols
in Microbiology (2006).
CFR Part 331, 9 CFR Part 121, 42 CFR Part 73,” July 5, 2013,
http://www.selectagents.gov/resources/Security_Guidance_v3-English.pdf.
2775 Ibid.
2776 United States Government Policy for Institutional Oversight of Life Sciences Dual Use Research of Concern. (2014)
Accessible at http://phe.gov/s3/dualuse/documents/durc-policy.pdf. Accessed on September 8, 2015.
2777 Ibid.

16.14 Laws, Guidance, Policies, Practices, and International Agreements on Biosafety
and Biosecurity
16.14.1 Types of Governing Instruments
16.14.1.1 Statutes2778,2779,2780
The federal statutes relevant for this assessment generally serve to prohibit certain activities, establish
criminal and civil penalties to deter such acts, and delegate regulatory authority to the executive branch.
Applicable statues usually do not specify functional operating requirements, which are left to the
regulatory authorities.2781
16.14.1.2 Regulations2782,2783,2784
Congress may empower an executive branch agency to establish and enforce regulations published in the
U.S. Code.2785 The regulations relevant for this assessment codify functional requirements while leaving
flexibility in implementation.2786 Federal regulations that apply to laboratory work are found throughout
the Code of Federal Regulations, from broad OSHA regulations on protecting workers from hazards, to
regulations on handling specific pathogens established by HHS and USDA.
Different regulations fall under different executive branch authorities. The three major federal regulatory
entities for GoF laboratories are the Occupational Safety and Health Administration (OSHA), the
Department of Health and Human Services (HHS), and the Department of Agriculture (USDA).2787 Other
agencies, such as the Department of Transportation, Department of Commerce, and the Environmental
Protection Agency are involved in smaller roles. Appendix V provides a detail list of all relevant laws and
guidance.
2778 Any “General and permanent” law passed by Congress is compiled into the U.S. Code.2778 The U.S. Code is the statutory
law of the country. In this analysis, However, it is “a rebuttable presumption that may be corrected” if one finds unrepealed
acts that are not reflected in the U.S. Code. Both the U.S. Code and acts of Congress published in the Federal Register are
considered.
2779 See: Richard J. McKinney, “Basic Overview on How Federal Laws Are Published, Organized and Cited,” FLICC Program
on Federal Legislative Research, January 2006, p.4 http://www.llsdc.org/assets/sourcebook/federal-laws.pdf.
2780 1 U.S.C. § 204 http://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title1-
section204&num=0&edition=prelimU.S. Government Publishing Office
http://www.gpo.gov/fdsys/browse/collection.action?collectionCode=PLAW
2781 For example, 18 U.S.C. § 175b leaves to regulation the designation of Federal Select Agents and Toxins. 18 U.S.C. §175b
http://www.gpo.gov/fdsys/pkg/USCODE-2011-title18/pdf/USCODE-2011-title18-partI-chap10-sec175b.pdf
2782 Federal regulations are compiled in the Code of Federal Regulations (CFR). In this report, federal regulations were retrieved
through the Electronic Code of Federal Regulations.
2783 U.S. Office of the Federal Register <http://www.ofr.gov/Catalog.aspx>.
2784 U.S. Government Publishing Office, “Electronic Code of Federal Regulations” <www.ecfr.gov>
2785 Richard J. McKinney, “Basic Overview on How Federal Laws Are Published, Organized and Cited,” FLICC Program on
Federal Legislative Research, January 2006, p.1 http://www.llsdc.org/assets/sourcebook/federal-laws.pdf.
2786 In effect, the regulations governing biological laboratories and their activities are not very prescriptive. Jennifer Gaudioso,
Susan A. Caskey, LouAnn Burnett, Erik Heegaard, Jeffery Owens, Philippe Stroot, “Strengthening Risk Governance in
Bioscience Laboratories,” Sandia National Laboratories, SAND2009-8070, December 2009, p.37,
http://www.biosecurity.sandia.gov/BioRAM/Biorisk%20Framework%20Report.pdf.
2787 U.S. Department of Health and Human Services (HHS), Public Health Emergency (PHE), “Biosafety and Biocontainment
FAQs,” http://www.phe.gov/s3/faqs/Pages/biosafety.aspx.

16.14.1.3 Guidance
Statutes and regulations are often very broad, and stress functionality rather than mandating means of
implementation.2788 Federal agencies frequently issue guidance to clarify regulations, establish best
practices, and provide additional optional recommendations to improve operations. Other organizations
like professional societies may also issue guidance and standards with practical recommendations.2789
Guidance documents often describe a baseline standard that all implementations of regulations must meet.
While some guidance documents provide optional recommendations, they often become viewed as de
facto requirements for regulatory compliance, certification, compliance with contracts, and/or liability
protection by the regulated community.
For instance, much of the guidance governing biosafety is enforced by OSHA through the authority
derived from a general employee hazards protection law. The Occupational Safety and Health Act of
1970 established the General Duty Clause, which required that employers to “furnish to each of his
employees employment and a place of employment which are free from recognized hazards that are
causing or likely to cause death or serious physical harm to his employees.”2790 Because of the broad all-
hazards and all-workplaces language of the Act, OSHA can incorporate guidelines provided by other
agencies, such as the CDC and NIH, effectively making compliance with the guidance mandatory.2791
HHS has integrated applicable laws, regulations, and best practices into the comprehensive guidance
document titled, Biosafety in Microbiological and Biomedical Laboratories (BMBL).2792 BMBL guidance
is broadly considered “the consensus code of practice for identifying and controlling biohazards,” and
adherence to the minimum requirements stated within the BMBL is enforced by regulators and all
research institutions.2793
16.14.1.4 Grants and Contracts
Guidance documents can also be enforced by making adherence to a contractual requirement or a term
and condition of award. NIH has employed this approach. NIH awardees are required, either through the
terms and conditions of an awarded grant or contractually, to meet worker health and safety standards.2794
U.S.-based institutions receiving NIH funding for any recombinant or synthetic nucleic acid research
must conduct biosafety risk assessment and risk management per the relevant NIH Guidelines for
Research Involving Recombinant or Synthetic Nucleic Acid Molecules.2795 Similarly, U.S.-based
2788 Jennifer Gaudioso, Susan A. Caskey, LouAnn Burnett, Erik Heegaard, Jeffery Owens, Philippe Stroot, “Strengthening Risk
Governance in Bioscience Laboratories,” Sandia National Laboratories, SAND2009-8070, December 2009, p.37,
http://www.biosecurity.sandia.gov/BioRAM/Biorisk%20Framework%20Report.pdf.
2789 For example: American Society of Heating, Refrigerating and Air-Conditioning Engineers, ASHRAE Laboratory Design
Guide, 1st edition (2002).
2790 Occupational Safety and Health Administration (OSHA), “Laboratory Safety Guidance,” OSHA 3404-11R, 2011, p.5,
https://www.osha.gov/Publications/laboratory/OSHA3404laboratory-safety-guidance.pdf.
2791 Ibid.
2792 U.S. Department of Health and Human Services, Biosafety in Microbiological and Biomedical Laboratories – Fifth Edition.
2794 National Institutes of Health, NIH Grants Policy Statement “4.1.12 Health and Safety Regulations and Guidelines,” October
2013, http://grants.nih.gov/grants/policy/nihgps_2013/nihgps_ch4.htm#health_safety_regulations.
2795 All NIH-funded projects using recombinant or synthetic nucleic acids and all projects at institutions that receive any NIH
funding, must conform to these guidelines.
National Institutes of Health, “NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules
(NIH Guidelines),” November 2013, <http://osp.od.nih.gov/sites/default/files/NIH_Guidelines_0.pdf>.

institutions receiving any federal funding for life science research are required to conduct dual use review
for experiments with certain agents and toxins, including HPAI and reconstructed 1918 influenza.2796
In general, the U.S. National Institutes of Health provides funding to researchers whose institutions
comply with applicable U.S. requirements per the grant award or contractual agreement, and the
researcher’s country, “providing the foreign requirements do not contradict U.S. laws.”2797 This includes
compliance with Select Agent Regulations, human subjects protections, animal care and use, recombinant
DNA guidelines, and other requirements as applicable. That said, an assessment of the landscape of
security governance and implementation at institutions outside the United States is extremely complex
because laws for securing pathogens differs significantly among countries. In addition, different countries
may categorize influenza, SARS-CoV, and MERS-CoV differently than the United States, which results
in different applicable country-specific laws and practices associated with research with these viruses.
Therefore, such an assessment would need to be country-specific and involve all relevant country
stakeholders (including law enforcement or security entities) to better understand the legal and practical
security environment in which U.S.-sponsored research is conducted.
16.14.1.5 International Obligations
Some federal statutes and regulations serve to implement obligations derived from international
agreements reached by the United States. For example, the U.S. has implemented its commitments under
the Biological Weapons Convention through legislation prohibiting biological weapons.2798,2799
Additionally, U.S. implementation of United Nations Security Council Resolution 1540 includes a variety
of legislative acts, executive orders, and regulations .2800,2801 Many international agreements require this
corresponding implementation to become practically enforceable in the U.S.; without thoughtfully crafted
implementing statutes and regulation, enforcing international commitments is difficult.2802,2803
Development, production, stockpiling, and use of biological information or material for biological
weapons purposes is outlawed by international law and is inconsistent with established international
norms. The 1925 Geneva Protocol bans the use of bacteriological and asphyxiating agents in war.2804 The
1972 Biological Weapons Convention (BWC) in essence bans the development, production, and
2797 NIH National Institute of Allergy and Infectious Diseases, “NIAID Select Agent Policy for Foreign Institutions Questions
and Answers,” May 13, 2015, http://www.niaid.nih.gov/researchfunding/qa/pages/selagentfor.aspx#standard. Accessed
November 11, 2015.
2798 Initial U.S. implementation was under the “Biological Weapons Anti-Terrorism Act of 1990” and has been updated under
the “Anti-Terrorism and Effective Death Penalty Act of 1996” and the “USA Patriot Act” of 2001.
U.S. Code, Title 18 Chapter 10-Biological Weapons Section 175, “Prohibitions with respect to biological weapons”
http://www.gpo.gov/fdsys/pkg/USCODE-2013-title18/pdf/USCODE-2013-title18-partI-chap10-sec175.pdf.;
2799 Text of the Biological Weapons Convention, 1972 http://www.state.gov/t/isn/bw/c48738.htm.
2800 United Nations Security Council, Resolution 1540 (2004)
http://www.un.org/en/ga/search/view_doc.asp?symbol=S/RES/1540%20(2004);
2801 Highlights include the Public Health Security and Bioterrorism Preparedness Response Act of 2002 modifying 18 USC
2283, the National Defense Authorization Act of 1995 (Public Law 103-337) and the Federal Select Agent Program. A
complete description of U.S. efforts under UNSCR 1540 can be found in October 11, 2013 letter from the U.S. to the U.N.
http://www.un.org/en/ga/search/view_doc.asp?symbol=S/AC.44/2013/17.
2802 U.S. Supreme Court, Medellín v. Texas, 552 U.S. 491 (2008) (No. 06-984).
2803 U.S. Supreme Court, Bond v. United States, 564 U.S. ___ (2014) (No. 12-158).
2804 Note that several countries at the time made treaty reservations reserving the right to retaliate in kind and/or limiting the ban
to cover only fellow Contracting Parties.
United Nations Office for Disarmament Affairs, “1925 Geneva Protocol: Protocol on the Prohibition of the Use in War of
Asphyxiating, Poisonous or Other Gases, and of Bacteriological Methods of Warfare,”
http://www.un.org/disarmament/WMD/Bio/1925GenevaProtocol.shtml.

stockpiling of biological weapons.2805 Most states have signed and ratified the treaty: the Convention has
173 States Parties and 9 signatories.2806 Only 14 UN-recognized states have not signed the Convention; of
these, only Israel and Angola have substantial armed forces.2807 The near-universality of the BWC means
that a strong case can now be made that a norm against the development, production, and stockpiling of a
biological weapon exists as a legally binding norm under international customary law.2808 However, the
Convention provides no mechanism for verification or enforcement, and some countries may be willing to
flout their obligations, as was done for instance by the Soviet Union.2809
International organizations, like the World Health Organization, may also issue guidance that
complements domestically-issued guidance, such as the Laboratory Biosafety Manual that complements
the BMBL.2810
16.14.1.6 Practice
Safety and security at high containment facilities are shared responsibilities among many stakeholders,
including the institution, Institutional Biosafety Committee, Institutional Review Entity, biosafety officer,
principal investigator, researchers, support staff, and law enforcement. Professional societies like the
American Biological Safety Association hold conferences that build a community of practice. Factors
such as the safety and security culture and personal relationships with emergency response personnel
drastically improve defenses but would not be captured by regulatory analysis.2811
In addition, institutions may implement measures beyond regulatory requirements. For instance, an
institution can decide to treat certain pathogens as if they were Tier 1 Select Agents for the purposes of
improved safety and security, going beyond what is required or broadly recommended in authoritative
guidance. Other institutions may implement additional physical security measures in order to safeguard
their personnel and laboratory space.
That said, implementation of biosafety and biosecurity measures varies across research institutions. On
one end of the spectrum are institutions that do not adequately comply with federal requirements and lose
funding or approval to conduct certain research. On the other end of the spectrum are institutions that go
well-above the minimum requirements for security as described in the governing documents. Therefore,
generalization of implementation across all research institutions is not appropriate and was not done in
this assessment.
2805 Convention on the Prohibition of the Development, Production and Stockpiling of Bacteriological (Biological) and Toxin
Weapons and on their Destruction,
http://www.unog.ch/80256EDD006B8954/%28httpAssets%29/C4048678A93B6934C1257188004848D0/$file/BWC-text-
English.pdf.
2806 United Nations Office at Geneva, The Biological Weapons Convention Implementation Support Unit, “Membership of the
Biological Weapons Convention,”
http://www.unog.ch/__80256ee600585943.nsf/%28httpPages%29/7be6cbbea0477b52c12571860035fd5c?OpenDocument&
ExpandSection=1#_Section1.
2807 The current list is: Angola, Chad, Comoros, Djibouti, Eritrea, Guinea, Israel, Kiribati, Micronesia (Federated States of),
Namibia, Niue, Samoa, South Sudan, Tuvalu.
2808 Nicholas A. Sims, “Legal Constraints on Biological Weapons,” Deadly Cultures: Biological Weapons since 1945, eds. Mark
Wheelis, Lajos Rózsa, Malcolm Dando (Cambridge: Harvard University Press, 2006), p. 331.
2809 For a comprehensive history of the Soviet biological weapons program, see: Leitenberg M, Zilinskas R, (2012) The Soviet
Biological Weapons Program: A History Cambridge: Harvard University Press.
2810 World Health Organization, Laboratory Biosafety Manual – Third Edition
http://www.who.int/entity/csr/resources/publications/biosafety/Biosafety7.pdf?ua=1.
2811 “Safety culture” as advocated by ABSA is also common in aviation and health care industries.
Felix Gmuender and Daniel Fischer, ABSA Conference Denver 2010, “Assessing Safety Culture in Biorisk Facilities”
http://www.absaconference.org/pdf53/Session12-Gmuender.pdf.

Because no systematic evaluation of all research institutions was possible for this assessment, measures
implemented in practice reflect those in use at the research institutions project staff visited during the
course of this assessment, which represent a total of 6 institutions conducting research involving
influenza, SARS-CoV, and/or MERS-CoV. Five institutions are subject to the U.S. Government’s pause
in funding and NIH’s “stop work” order of GoF research. One institution that also has received a “stop
work” order does not conduct any research with Biological Select Agents and Toxins.
16.14.2 Laws, International Agreements, and Guidance Documents
The following tabular list of laws, international agreements (including treaties and other international
obligations), and guidance documents on biosafety and biosecurity was compiled as part of the above
analysis on the policies and practices governing U.S. laboratories in the biosafety and biosecurity spheres.
The table provides the relevant item name as well as a hyperlink to allow retrieval of the item. For each
item, the table contains a short summary highlighting the relevant aspects of the item for the current
report. Each item is assigned a subjective relevance score to indicate how applicable the item was to the
assessment in the current report (low relevance; 5 high relevance). Where applicable, the item is classed
as “safety,” “security,” or “safety and security” -oriented, depending on the motivation behind the item.
Finally, each item is given a topic classification based on the safety/security functions the item performs.
The numbers are as follows:
1 Personnel surety
2 Physical/electronic access control
3 Inventory/accountability
4 Storage
5 Transfer, shipment, chain-of-custody
6 Surveillance and monitoring
7 Malicious actor detection
8 Incident Reporting
9 Emergency Response
0 Research Plan
A Waste disposal
Items having several different functions are given a combined number, in order. So for instance an item
like 29 CFR 1910.1201 that deals with Inventory/accountability issues and transfer, shipment, and chain-
of-custody issues, is assigned the number 35.

Table 16.5. Laws, regulations, international agreements, and guidance documents on biosafety and biosecurity
Item Item Type Date Safety/ Safety/ Subjective Summary Link

Security Security Relevance
? Functions (1=low
5=high)
Export Administration Regulations (Parts 730 to

780) and Additional Protocol Regulations (Part 781
to 799). The Commerce Control List under 5 CFR
Federal
5 CFR 730-799 Security 5 0 738 regulates inter alia exports of pathogens by Link
Regulations
potentially requiring an export license depending on
the pathogen and its destination (essentially for
national security / global security reasons).
7 CFR 330: Code of

Federal Regulations,
Federal
Title 7 "Agriculture," 0 Regulations on plant pests Link
Regulations
Part 330 "General
Provisions"
7 CFR 331: Code of [PPQ SELECT AGENTS] Implementation of the

Federal Regulations, Agricultural Bioterrorism Protection Act of 2002
Title 7 "Agriculture," (alongside 9 CFR 121). Note that the safety and
Federal
Part 331 "Possession, 0 security regulations under 7 CFR 331 are Link
Regulations
Use, and Transfer of functionally equivalent to those laid out for USDA
Select Agents and Select Agents and CDC Select Agents, albeit for
Toxins" different pathogens.
7 CFR 331.3: Code of [PPQ SELECT AGENTS] Lists plant pathogens

Federal Regulations, classed as PPQ Select Agents and regulated by 7
Federal
Title 7 "Agriculture," 0 CFR 331. PPQ is the "Plant Protection and Link
Regulations
Part 331.3 "PPQ select Quarantine Programs of the Animal and Plant Health
agents and toxins" Inspection Service."


? Functions (1=low
5=high)
[USDA SELECT AGENTS] Implementation of the

9 CFR 121: Code of
Agricultural Bioterrorism Protection Act of 2002
(alongside 7 CFR 331). The safety and security
Title 9 "Animals and
Federal regulations under 9 CFR 121 are functionally
Animal Products," Part Link
Regulations equivalent to those laid out for PPQ Select Agents
121 "Possession, Use,
and CDC Select Agents, albeit for different
and Transfer of Select
pathogens. Influenza is a Veterinary Services Select
Agents and Toxins"
Agent (VS Select Agents; i.e. USDA Select Agents).
A permit issued by the USDA Secretary is required

to transport any organisms or vectors across
9 CFR 122: Code of
state/territory/district of Columbia lines or to import
them into the United States, unless a permit has
Title 9 "Animals and Federal
already been granted or the organism was produced Link
Animal Products," Part Regulations
at an establishment licensed under 9 CFR 102. The
122 "Organisms and
rest of 9 CFR 122 covers the permit application
Vectors"
process and the suspension or revocation of permits
process.
Federal Requirements and standards for Accredited

9 CFR 161 2 Link
Regulations Veterinarians.
Commerce and foreign trade regulations.

15 CFR Parts 730-774:
Regulations 15 CFR 710 to 721 implement the
Code of Federal
Chemical Weapons Convention (CWC), see entry
Regulations, Title 15 Federal
0 below. Regulation 15 CFR 744.6 calls for a Bureau Link
"Commerce and Regulations
of Industry and Security (BIS) license to export or
Foreign Trade," Parts
transfer any item that could be used in development
730-774
of a biological weapon.


? Functions (1=low
5=high)
The set of regulations under 21 CFR 58 cover

laboratory practices for nonclinical studies, but apply
to testing facilities that do safety tests on test articles,
and not clinical studies or field trials in animals.
Therefore, its relevance for biological research
laboratories considered in the current study is low.
Two parts (58.81 and 58.90) are flagged here for
comparative purposes. Part 58.81 "Standard
Operating Procedures" sets the requirements for a
21 CFR 58: Code of standard operating plan, which must list instructions
Federal Regulations, for a large number of common laboratory tasks
Title 21 "Food and (detailed in the code). For instance, instructions on
Federal
Drugs," Part 58 "Good None 0 how to conduct animal room preparation; on how to Link
Regulations
Laboratory Practice for ensure animal care; on the "placement, transfer, and
Nonclinical Laboratory identification of animals"; on how to handle animals
Studies" "found moribund or dead during a study"; and on the
"maintenance and calibration of equipment." Animal
care regulations are themselves detailed in Part 58.90
"Animal Care." These regulations include the
isolation and health assessment of newly received
animals, and the suitable identification of warm-
blooded animals that are not suckling rodents that
must be manipulated for "an extended period of
time" or that must be removed from and returned to
their cages for any reason (including cleaning).


? Functions (1=low
5=high)
The labor regulations under 29 CFR 1910 apply to

"workplaces in general industry" apart from mobile
workplaces (vehicles, vessels). 29 CFR 1910.38
specify regulations on emergency action plans
required by OSHA. Such emergency action plans
must include procedures: for reporting emergencies
29 CFR 1910.38: Code (such as fires); for emergency evacuation; for
of Federal Regulations, 7 employees who must remain "to operate critical
Federal
Title 29 "Labor," Part November 3 plant operations before they evacuate"; for Link
Regulations
1910.38 "Emergency 2002 employees performing rescue or medical duties; for
action plans" ensuring all employees are accounted for after an
evacuation; and contact information for employees
that can be reached by other employees for
information on the emergency action plan. There
must be an alarm system to warn employees, and
employees must be trained in assisting "in a safe and
orderly evacuation of other employees."


? Functions (1=low
5=high)
The hazardous waste operations and emergency

response regulations under 29 CFR 1910.120
explicitly apply to biological agent hazards, as a
"hazardous substance" includes "any biological
agent and other disease-causing agent which after
release into the environment" may cause adverse
effects in individuals (including "death, disease,
behavioral abnormalities"). The regulation stipulates
that employees must have a written safety and health
plan for employees for normal facility work with
hazardous wastes, and a site-specific safety and
29 CFR 1910.120:
health plan for such tasks. Mandated elements of a
Code of Federal
site-specific plan of particular relevance to this
Regulations, Title 29
Federal report include the requirement for written "lines of
"Labor," Part 1910.120 4 Link
Regulations authority, responsibility, and communication" the
"Hazardous waste
provision of personal protective equipment (PPEs)
operations and
and when necessary decontamination showers, and
emergency response"
the setup of a medical surveillance program. The
facility must have developed and communicated
decontamination procedures to employees. The
facility must also have an emergency response plan
which must include inter alia: emergency alerting
procedures, the provision of PPEs and emergency
equipment, the provision of emergency medical
treatment and first aid, and specific decontamination
procedures that are not covered by the safety and
health plan. This emergency response plan must be
"rehearsed regularly."


? Functions (1=low
5=high)
29 CFR parts
1910.132-138, Annex
A and Annex B: Code
of Federal Regulations,
The regulations set requirements for selecting,
Title 29 "Food and Federal
1 1 providing, maintaining, and replacing personal Link
Drugs," Part 1910.132- Regulations
protective equipment.
138, Annex A and
Annex B "Subpart I-
Personal Protective
Equipment
These regulations apply to all occupational exposure

to "human blood, human blood components, and
products made from human blood," to "pathogenic
29 CFR 1910.1030: Last
microorganisms that are present in human blood and
Code of Federal amended
can cause disease in humans" (such as Hepatitis B
Regulations, Title 29 3 April
Federal virus and human immunodeficiency virus), and a
"Labor," Part 2012; 5 Link
Regulations defined list of human bodily fluids, unfixed human
1910.1030 initial 6
tissue or organs, and "HIV-containing cell or tissue
"Bloodborne December
cultures, organ cultures, and HIV- or HBV-
pathogens" 1991
containing culture medium or other solutions; and
blood, organs, or other tissues from experimental
animals infected with HIV or HBV."
Last
29 CFR 1910.1200: amended
Code of Federal 8
These regulations on hazard communications
Regulations, Title 29 Federal February
0 explicitly do not apply to "biological hazards" as per Link
"Labor," Part Regulations 2013;
6), xii).
1910.1200 "Hazard initial 9
communication" February
1994


? Functions (1=low
5=high)
Relevant portion: any individual receiving a

hazardous material shipment that must be marked,
29 CFR 1910.1201:
labelled, or placarded, must retain the "markings,
Code of Federal
labels and placards" required under U.S. Department
Federal 19 July of Transportation's Hazardous Materials Regulations
"Labor," Part 35 1 Link
Regulations 1994 (49 CFR 171 through 180). "For non-bulk packages
1910.1201 "Retention
which will not be reshipped, the provisions of this
of DOT markings,
section are met if a label or other acceptable marking
placards and labels"
is affixed in accordance with the Hazard
Communication Standard (29 CFR 1910.1200)."
29 CFR 1910.1450: Last

Code of Federal amended
Regulations, Title 29 22 The regulations promulgated under this part regard
"Labor," Part Federal January occupational chemical exposure hazards. However,
2 Link
1910.1450 Regulations 2013; the part includes a section "I. Laboratory Security"
"Occupational exposure initial 31 which is also applicable for biosecurity assessments.
to hazardous chemicals January
in laboratories" 1990


? Functions (1=low
5=high)
Establishes regulations regarding safety of

employees during construction that are extremely
similar to those under 29 CFR 1910. Of note,
1926.65 "Hazardous Waste Operations and
29 CFR 1926: Code of
Emergency Response" establishes regulations for
"emergency response operations for releases of, or
Title 29 "Labor," Part Federal
0 substantial threats of releases of, hazardous Link
1926 "Safety and Regulations
substances without regard to the location of the
Health Regulations for
hazard." A hazardous substance is defined as a
Construction"
"substance which, by reason of being explosive,
flammable, poisonous, corrosive, oxidizing,
irritating, or otherwise harmful, [are] likely to cause
death or injury."
39 CFR 20: Code of

International mail manual. The international mail
Title 39 "Postal Federal
5 2 manual itself has regulations (Section 601.10.17) on Link
Service," Part 20 Regulations
transporting infectious substances through USPS.
"International Postal
Service"
40 CFR parts 150-189:

Regulations on pesticides. 40 CFR 160 establishes
Code of Federal
regulations for "good laboratory practice standards"
Federal for conducting studies "that support or are intended
"Protection of the None 0 Link
Regulations to support applications for research or marketing
Environment," Parts
permits for pesticide products regulated by the
150-189 "Subchapter
EPA."
E- Pesticide Programs"


? Functions (1=low
5=high)
Regulations governing the importation of “infectious

biological agents, infectious substances, and vectors”
into the U.S. from abroad. Such activities are
prohibited without a permit. The CDC issues permits
42 CFR 71.54: Code of
which then detail the specific requirements and
conditions placed on the sample (which can include
Title 42 "Public
restrictions on intra-state transfer once in the U.S.).
Health," Part 54 4
Federal The importer must implement “biosafety measures
"Import regulations for February 5 2 Link
Regulations commensurate with the hazard posed by the
infectious biological 2013
infectious biological agent, infectious substance,
agents, infectious
and/or vector to be imported, and the level of risk
substances, and
given its intended risk.” The importer must also
vectors"
“help ensure” that the shipper complies with all
applicable legal requirements “concerning the
packaging, labeling, and shipment of infectious
substances.”
42 CFR 72 Federal
None 0 Not valid anymore [Reserved]
[RESERVED] Regulations
42 CFR 73.0: Code of Amended

[CDC SELECT AGENTS] Possession of SARS-
Federal Regulations, 4
CoV, Lujo virus, Chapare virus must be reported to
Title 42 "Public December
Federal CDC on or before December 2012. Compliance with
Health," Part 73 "Select 2012; 3 5 Link
Regulations the rest of 42 CFR 73 is required for new registrees
Agents", subsection 0 original 5
by April 3, 2013 and already registered possessors
"Applicability and October
by December 4, 2012.
related requirements" 2012


? Functions (1=low
5=high)
[CDC SELECT AGENTS] Defines the CDC's HHS

select agents and toxins, and identifies certain of
42 CFR 73.3: Code of Last
these select agents and toxins as Tier 1 select agents
Federal Regulations, amended
and toxins. SARS-CoV and "Reconstructed
Title 42 "Public 12 May
Federal replication competent forms of the 1918 pandemic
Health," Part 73 "Select 2014; 5 Link
Regulations influenza virus containing any portion of the coding
Agents", subsection 3 original
regions of all eight gene segments (Reconstructed
"HHS select agents and 18 March
1918 Influenza virus)" are both select agents, but are
toxins" 2005
not Tier 1 select agents. MERS-CoV is as of August
2015 not a select agent.
[CDC SELECT AGENTS] Defines overlap select

agents and toxins, and identifies certain of these
overlap select agents and toxins as Tier 1 overlap
Federal select agents and toxins. Overlap agents and toxins
Health," Part 73 "Select 2014; 1 Link
Regulations are those subject to regulation by both CDC and
APHIS. SARS-CoV, MERS-CoV, and influenza are
"Overlap select agents 18 March
as of August 2015 not overlap select agents and
and toxins" 2005
toxins.


? Functions (1=low
5=high)
[CDC SELECT AGENTS] Possession, use, or

transfer of HHS select agent or toxin requires a
certificate of registration issued by the HHS
Secretary (exceptions exist as listed in 73.5 for
clinical and diagnostic labs shipping select agent
pathogens/toxins in specimens for diagnosis or
42 CFR 73.7: Code of verification or for proficiency testing, as well as for
Federal Regulations, products given specific exemptions. These are
Title 42 "Public irrelevant here.) The Attorney General must do a risk
Health," Part 73 "Select Federal assessment before granting registration and the HHS
35 5 Link
Agents", subsection 7 Regulations Secretary needs to base the decision to grant
"Registration and registration on this assessment. This assessment is
related security risk based on information provided by those seeking
assessments" registration through APHIS/CDC Form 1, and can
also be based on inspection or submission of
additional documents prepared under 42 CFR 73
requirements (such as the security plan). Certificate
of registration is valid for a maximum of 3 years.
[CDC SELECT AGENTS] Those registering need
to designate a Responsible Official.

[CDC SELECT AGENTS] Provides clauses for
denying, revoking, or suspending a certification of
Title 42 "Public
registration. If a certification of registration is
Health," Part 73 "Select Federal
1A 1 revoked or suspended, all work with select agents Link
Agents", subsection 8 Regulations
and toxins must stop. The select agents and toxins
"Denial, revocation, or
must be safeguarded, and if HHS requests it they
suspension of
must be disposed of as requested.
registration"


? Functions (1=low
5=high)
[CDC SELECT AGENTS] Sets requirements and

duties of the Responsible Official. They must carry
out an annual documented inspections of registered
laboratories that stored or used pathogens. They
42 CFR 73.9: Code of Amended must also report the identification of select agents
Federal Regulations, 5 October and toxins contained in diagnosis or verification
Title 42 "Public Federal 2012; specimens within seven calendar days after
3 3 Link
Health," Part 73 "Select Regulations original identification for SARS-CoV and reconstructed
Agents", subsection 9 18 March influenza virus in diagnosis or verification
"Responsible Official" 2005 specimens and within 90 days for proficiency testing
specimens; the reporting is done through
APHIS/CDC Form 4 and a copy of the form must be
kept for three years (clauses are more stringent for
some pathogens; requires telephone call reporting).


? Functions (1=low
5=high)
[CDC SELECT AGENTS] An individual's access

to a select agent or toxin must be pre-approved by
the HHS Secretary or HHS Administrator, following
a security risk assessment conducted by the Attorney
General. Access is defined as the possession of a
select agent or toxin (such as the ability to use,
manipulate, carry) or the ability to gain possession of
Amended a select agent or toxin. The approval is valid for a
Title 42 "Public
5 October maximum of three years. The individual must have
Health," Part 73 "Select
Federal 2012; Safety & "the appropriate education, training, and/or
Agents", subsection 10 1 5 Link
Regulations original Security experience to handle or use such agents or toxins."
"Restricting access to
18 March The regulation provides clauses so that HHS can
select agents and
2005 deny, limit, or revoke an individual's access approval
toxins; security risk
for safety or security reasons. Further, should an
assessments"
individual's access to select agents or toxins be
terminated by their entity (not HHS), the
Responsible Official must "immediately notify"
CDC or APHIS and must present the reason(s)
behind the decision (for instance, a researcher
changing laboratories).


? Functions (1=low
5=high)
[CDC SELECT AGENTS] A security plan must be

developed and implemented by those registering to
meet CFR 73 regulations that is "sufficient to
safeguard the select agent or toxin against
unauthorized access, theft, loss, or release." The
security plan must address 10 specific security-
related topics (see link), which can be summarized
as covering procedures for routine operations
(cleaning, maintenance, repairs), for facility security
Last
42 CFR 73.11: Code of (such as establishing a minimum of three security
amended
Federal Regulations, barriers, setting reporting requirements,
12 May
Title 42 "Public Federal 12345678 implementing inventory control, securing storage of
2014; Security 5 Link
Health," Part 73 "Select Regulations 9 select agents and toxins, following cyber-security
original
Agents", subsection 11 measures, and inspecting suspicious packages
18 March
"Security" outside of areas where select agents and toxins are
2005
used or stored, to prompt first response of security
forces), for transfers of select agents or toxins
(shipping to another entity, intra-entity transfers), for
emergency response (removing unauthorized or
suspicious personnel, responding to exposure of
animals or plants, to address security compromises
such as lost keys, to communicate with law
enforcement), and to provide for personnel training
and personnel protocols (reporting channels).


? Functions (1=low
5=high)
[CDC SELECT AGENTS] A biosafety plan must

42 CFR 73.12: Code of Amended be developed and implemented by those registering
Federal Regulations, 5 October to meet CFR 73 regulations. It must include
Title 42 "Public Federal 2012; descriptions of the biosafety and containment
Safety 09 5 Link
Health," Part 73 "Select Regulations original procedures for the select agent or toxin as well as
Agents", subsection 12 18 March "any animals (including anthropods) or plants
"Biosafety" 2005 intentionally or accidentally exposed to or infected
with a select agent."
[CDC SELECT AGENTS] Restricts the conduct of

certain experiments and the possession of results
from said restricted experiments, unless approved
and conducted as requested by the HHS Secretary.
The experiments restricted are those that: "involve
the deliberate transfer of, or selection for, a drug
Federal Safety & resistance trait to select agents that are not known to
Health," Part 73 "Select 2014; 0 5 Link
Regulations Security acquire the trait naturally, if such acquisition could
compromise the control of disease agents in humans,
"Restricted 18 March
veterinary medicine, or agriculture" or "experiments
experiments" 2005
involving the deliberate formation of synthetic or
recombinant DNA containing genes for the
biosynthesis of select toxins lethal for vertebrates at
an LD[50] <100".


? Functions (1=low
5=high)
[CDC SELECT AGENTS] A written incident

response plan must be developed based on a site-
specific risk assessment. This plan must be kept
available for review by "employees" (no further
explanation), and the incident response plan must be
42 CFR 73.14: Code of Amended exercised at least yearly. There are specific
Federal Regulations, 5 October additional requirements for facilities with Tier 1
Title 42 "Public Federal 2012; Safety & select agents. The incident response plan must, inter
5 Link
Health," Part 73 "Select Regulations original Security alia, "fully describe the entity's response procedures
Agents", subsection 14 18 March for the theft, loss, or release of a select agent or
"Incident response" 2005 toxin; inventory discrepancies; security breaches
(including information systems); severe weather and
other natural disasters; workplace violence; bomb
threats and suspicious packages; and emergencies
such as fire, gas leak, explosion, power outage, and
other natural and man-made events."
[CDC SELECT AGENTS] Those registering to

meet CFR 73 regulations must provide training on
biosafety, security, and incident response for
personnel to be working with select agents or toxins,
or that will enter areas where select agents or toxins
are stored or handled. The training must be done
Title 42 "Public Federal 5 October Safety &
58 5 before that person is granted access by HHS. Link
Health," Part 73 "Select Regulations 2012 Security
Refresher training must be done annually, or
Agents", subsection 15
whenever there is a "significant" amendment to
"Training"
biosafety, security, or incident response plans.
Training must be logged. Facilities holding Tier 1
select agents must in addition conduct yearly
specific annual insider threat awareness training.


? Functions (1=low
5=high)
[CDC SELECT AGENTS] Regulations for the

transfer of select agents or toxins, which can only be
conducted between individuals or entities registered
to possess select agents or toxins. CDC or APHIS
approval is required before a transfer unless the
42 CFR 73.16: Code of Select Agent is contained in a specimen for
Federal Regulations, proficiency testing, in which case CDC or APHIS
Title 42 "Public Federal Safety & must simply be informed at least 7 days prior to the
5 3 Link
Health," Part 73 "Select Regulations Security transfer (unless the transferors are both under the
Agents", subsection 16 same entity for the registration). Authorization for
"Transfers" transfer is sought by submitting APHIS/CDC Form
2. If the select agent or toxin has not been received
within 48 hours after the slated delivery date, or if
the package is damaged "to the extent that a release
of the select agent or toxin may have occurred," the
receiver must immediately notify CDC or APHIS.
[CDC SELECT AGENTS] Records of the name

and characteristics ("strain, GenBank Accession
number, etc."), the quantity acquired and date and
42 CFR 73.17: Code of source of acquisition, the storage location, the
Federal Regulations, movement in-and-out of storage of the sample and
Title 42 "Public Federal Safety & the individual(s) who moved the sample, intra-entity
34 5 Link
Health," Part 73 "Select Regulations Security transfer records, external transfer records, and a list
Agents", subsection 17 of all animals and plants intentionally or accidentally
"Records" exposed to or infected with a select agent must be
kept for any select agent "held in long-term storage."
Similar regulations are established for toxins
(omitted here).


? Functions (1=low
5=high)
[CDC SELECT AGENTS] Allows the HHS

secretary to inspect without prior notification any
site where activities regulated by 42 CFR 73 take
Title 42 "Public Federal
3 1 place, and will be allowed to inspect and copy Link
Health," Part 73 "Select Regulations
relevant records. The HHS secretary can conduct an
Agents", subsection 18
inspection prior to issuing a certificate of registration
"Inspections"
(As also noted in CFR 73.7).
[CDC SELECT AGENTS] Requires immediately

reporting the theft or loss of a select agent or toxin to
CDC or APHIS and to "appropriate Federal, State, or
local law enforcement agencies," without exception
(for instance regardless of whether the select agent
or toxin is then identified, or the responsible parties
Title 42 "Public
Federal Safety & found). A completed APHIS/CDC Form 3 must then
Health," Part 73 "Select 89 5 Link
Regulations Security be submitted within seven calendar days. Also
Agents," subsection 19
requires immediately reporting the" release of an
"Notification of theft,
agent or toxin causing occupational exposure or
loss, or release"
release of a select agent or toxin outside of the
primary barriers of the biocontainment area" to CDC
or APHIS. A completed APHIS/CDC Form 3 must
then be submitted within seven calendar days.


? Functions (1=low
5=high)
171.15 requires individuals with physical possession

49 CFR 171.15 and
of a hazardous material to notify by telephone the
171.16: Code of
National Response Center "as soon as is practical but
Last no later than 12 hours after" when certain types of
Title 49
amended incidents involving the hazardous material occurs.
“Transportation,” Part
20 July This specifically includes "fire, breakage, spillage, or
171.15 “Immediate Federal
2011; Safety 59 2 suspected contamination" involving an infectious Link
notice of certain Regulations
initial 3 substance other than regulated medical waste. They
hazardous materials
December are then also required to fill out a detailed incident
incidents,” and Part
2003 report (Hazardous Materials Incident Report on DOT
171.16 “Detailed
Form F 5800.1 (01/2004)) within 30 days of the
hazardous materials
incident. The report parameters are detailed in
incident reports”
171.16.
49 CFR 172.802: Code

Requires a security plan for transportation of certain
hazardous materials, but Division 6.2 materials
Title 49 Federal 16 April
Security 5 0 (infectious substances) are not one of the listed Link
"Transportation," Part Regulations 2008
hazardous materials covered by this set of
802 "Components of a
regulations.
security plan"


? Functions (1=low
5=high)
49 CFR Parts 171 to 180 regulate the transport of

hazardous materials. Under 49 CFR 173.134,
infectious substances are called "Division 6.2"
materials, and are defined as materials "known or
reasonably expected to contain a pathogen" (except
neutralized or inactivated materials). Infectious
49 CFR 173.134: Code
Last substances are then categorized as either Category A
amended or Category B. Category A is for an "infectious
Title 49
11 March substance in a form capable of causing permanent
“Transportation,” Part Federal
2013; 5 2 disability or life-threatening or fatal disease in Link
173.134 “Class 6, Regulations
initial 14 otherwise healthy humans or animals when exposure
Division 6.2-
August to it occurs", whereas Category B is for an infectious
Definitions and
2002 substance that is not in such a form. Classification of
exceptions”
an infectious substance as Category A or B "must be
based on the known medical history or symptoms of
the source patient or animal, endemic local
conditions, or professional judgment concerning the
individual circumstances of the source human or
animal."
Last
49 CFR 173.196: Code Regulations for the shipment of Category A
amended
of Federal Regulations, substances are given under 49 CFR 173.196. The
7 January
Title 49 Federal triple-packing requirement is detailed. The primary
2013; Safety 5 2 Link
“Transportation,” Part Regulations receptacle must be capable of resisting given
initial 14
173.196 “Category A pressure and temperature ranges without leaking.
August
Infectious substances” See link for details.
2002


? Functions (1=low
5=high)
Regulations for the shipment of Category B

substances are given under 49 CFR 173.199.
Category B substances must be triple-packed (two
Last
49 CFR 173.199: Code receptacles and a rigid outer packaging), and the
amended
of Federal Regulations, requirements for each layer of packaging are laid out
8 January
Title 49 Federal (see link). In case of transportation by aircraft, the
2015; Safety 5 2 Link
“Transportation,” Part Regulations package is inspected for leakage; if leakage is
initial 14
173.199 “Category B detected, then the cargo compartment must be
August
Infectious substances” disinfected. The regulation has a training component
2002
requiring that "each person who offers or transports"
a Category B infectious substance know of the
requirements under this regulation section.
49 CFR 178.609: Code Last

of Federal Regulations, amended
Title 49 7 Provides regulations on the test standards for
“Transportation,” Part Federal Septembe packaging materials required for infectious
Safety 1 Link
178.609 “Test Regulations r 2004; substances (and hence, for Category A and Category
equipment for initial 21 B agents).
packagings for December
infectious substances” 1990
Last
amended
6 October This law empowered the Secretary of Labor to enact
"Occupational Safety Federal
1992; Safety 1 regulations on occupational safety of employees Link
and Health Act" Laws
initial 29 engaged in hazardous waste operations.
December
1970


? Functions (1=low
5=high)
The part most relevant for this report is Title II,

“Enhancing Controls on Dangerous Biological
Agents and Toxins.” The sections under this Title
amended the Antiterrorism and Effective Death
Penalty Act of 1996 controls on biological agents
“Public Health Security
Safety (see below). In particular, the act adds text to require
and Bioterrorism Federal 12 June
and 5 the Secretary to enact regulation that became the Link
Preparedness and Laws 2002
Security Select Agents regulations, i.e. on possession (and on
Response Act of 2002”
barring possession from restricted persons),
transfers, and incident reporting for dangerous
pathogens per the dangerous pathogens list (the
Select Agents, although not called as such in the
act).
Title II of the Public Health Security and

… “Agricultural Bioterrorism Preparedness and Response Act of
Bioterrorism Protection 2002, “Enhancing Controls of Dangerous Biological
Act of 2002,” within Agents and Toxins,” has a Subtitle B cited as the
the “Public Health “Agricultural Bioterrorism Protection Act of 2002.”
Security and This text created the Secretary of Agriculture’s
Bioterrorism Select Agents list and associated regulations for
Preparedness and pathogens with “the potential to pose a severe threat
Response Act of 2002” to animal or plant health, or to animal or plant
products.”
Updates 18 USC 175 (see below) under Section 817,

26
Federal notably by adding a definition of “restricted persons”
“USA PATRIOT Act” October Security 1 Link
Laws and making it so that such persons are prohibited
2001
from having access to Select Agents.


? Functions (1=low
5=high)
Codifies the BWC's Articles I and III into U.S.

Last national law by criminalizing the "development,
18 USC 175: amended production, transfer, acquisition, retention, or
"Prohibitions with Federal 12 June possession of any biological agent, toxin, or delivery
Security 1 Link
Respect to Biological Laws 2002; system for other than prophylactic, protective, bona
Weapons" initial 22 fide research, or other peaceful purposes." The law’s
May 1990 official short-hand is the 'Biological Weapons Anti-
Terrorism Act of 1989'.
Contains text on enhanced control over biological

agents, as well as enhanced penalties for
unauthorized possession of biological agents.
Antiterrorism and
Federal 24 April Creates the first Select Agents list (although not
Effective Death Penalty Security 3 Link
Laws 1996 called as such in the act) by requiring “the Secretary
Act of 1996
[…to] establish and maintain a list of each biological
agent that has the potential to pose a severe threat to
public health and safety.”
Executive Order 13546,

“Optimizing the
Safety
Security of Biological Executive 2 July Established the risk-based tiering of the Select
and 5 Link
Select Agents and Order 2010 Agents, into Select Agents and Tier 1 Select Agents.
Security
Toxins in the United
States”
Created a Working Group on Strengthening the

Executive Order 13486,
Biosecurity of the United States within the
“Strengthening Executive 9 January
Security 0 Department of Defense with a mandate to review the Link
Laboratory Biosecurity Order 2009
effectiveness of relevant laws, regulations, guidance,
in the United States”
and practices.


? Functions (1=low
5=high)
The Biological Weapons Convention's Article III

reads: "Each State Party to this Convention
undertakes not to transfer to any recipient
whatsoever, directly or indirectly, and not in any
way assist, encourage, or induce any State, group of
States or international organizations to manufacture
Signed 10 or otherwise acquire any of the agents, toxins,
April weapons, equipment or means of delivery specified
1972; in Article I of the Convention." The treaty text does
Biological Weapons International
entered Security 1 not contain steps that State Parties must take to be in Link
Convention Treaty
into force compliance with this Article. In practice, States
26 March Parties such as the U.S. have passed national laws
1975 and established regulations ("National
Implementation") that restrict access to dangerous
pathogens and criminalize unauthorized access.
More specifically, the U.S. Biological Weapons
Anti-Terrorism Act of 1989 (enacted 1990, amended
1996) provided for the BWC's implementing (see its
entry above).
Signed
The Chemical Weapons Convention (CWC)
13, 1993;
regulates inter alia toxin production and stockpiling.
Chemical Weapons International entered
Security 0 The U.S. has established a series of regulations Link
Convention Treaty into force
under 15 CFR 710 to 721 for the national
29 April
implementation of the CWC.
1997


? Functions (1=low
5=high)
The Australia Group is an informal group of states

(including the U.S.) that seek to harmonize export
controls on chemical and biological agents and
equipment. The Australia Group maintains Common
Control Lists that are meant as guides of what to
restrict through state-level national export control
Informal
laws and regulations. The Common Control List
Australia Group international Security 0 Link
regarding Human and Animal Pathogens and Toxins
grouping
for Export Control includes "SARS-CoV-related
coronavirus," "Avian influenza viruses of high
pathogenicity" (as defined by WHO, the EU, or
competent national regulatory bodies) and
"Reconstructed 1918 influenza virus," as well as
certain genetic elements thereof.
United Nations Security Council Resolution

(UNSCR) 1540 is a legally-binding resolution on all
UN Member States that requires these states to
deploy measures against biological, chemical, and
Legally-
April nuclear weapons proliferation, "including
UNSCR 1540 binding Security 0 Link
2004 appropriate laws and regulations to control export,
UNSCR
transit, trans-shipment and re-export." Of relevance
here is the establishment of dangerous pathogen
export controls prompted/assisted by the 1540
Committee’s work.


? Functions (1=low
5=high)
Urges member states to take a number of measures

under the "Enhancement of Laboratory Biosafety"
rubric, including to: review lab safety protocols;
World Health implement programs to promote biosafety for safe
Int'l 25 May
Assembly Resolution Safety 1 handling and transport; develop national Link
Agreement 2005
58.29 (2005) preparedness plans and plans to enhance lab
compliance with biosafety guidelines for lab
practices; and to facilitate international access to lab
biosafety equipment (such as PPEs).
IHR is legally binding for all WHO member states,

International Health since WHA 58.3 is a World Health Assembly
Regulations, World Int'l resolution that adopts the IHR. The regulations are
May 2005 Safety 0 Link
Health Assembly Agreement "to prevent, protect against, control and provide a
Resolution 58.3 (2005) public health response to the international spread of
disease."
The Organisation for Economic Co-operation and

OECD Best Practice
Development (OECD) has issued the "OECD Best
Guidelines for Int’l March
Practice Guidelines for BRCs [Biological Resource Link
Biological Resource Agreement 2007
Centers]." OECD member countries (including the
Centers
U.S.) agreed to these guidelines in March 2007.
Army Regulations regarding biosafety/biosecurity.

Chapter 2 details the personnel reliability program
process to be followed. Chapter 3 details pathogen
and toxin control and inventory management.
Military 28 July Safety & Chapter 4 details Army procedures for transport.
Army Regulations 50-1 3 Link
regulations 2008 Security Chapter 5 details the occupational health program
process to be followed. Chapter 6 details the security
program process to be followed. Chapter 7 lays out
the incident response process. Chapter 8 details the
surety program evaluations procedures.


? Functions (1=low
5=high)
42 CFR 73.11 notes that those designing a security

Security Guidance for
5 July plan "should consider" this document. The document
Select Agent or Toxin Guidance Security 5 Link
2013 provides guidance on how to implement the required
Facilities
security aspects of the Select Agents regulations.
Last
revised 16
Guidance on the April Safety
Guidance on proper storage and inventory
Inventory of Select Guidance 2015; and 34 5 Link
management for select agents.
Agents and Toxins initial 12 Security
October
2012


? Functions (1=low
5=high)
NIH Guidelines on the conduct of recombinant or

synthetic nucleic acid molecule research. All NIH
funded projects on the topic, as well as non-NIH
funded projects on the topic carried out at or
supported by institutions that receive NIH funding,
must conform to these guidelines (p.11). Influenza
generated by recombinant or synthetic methods are
to be run under the biosafety level that would be
used if dealing with the virus from which the
Amended
NIH Guidelines for majority source of segments came from (p.21). BSL-
6
Research Involving 3 enhanced containment is to be used for all
November
Recombinant or Guidance 5 influenza viruses “containing genes or segments Link
2013;
Synthetic Nucleic Acid from 1918-1919 H1N1 (1918 H1N1), human H2N2
Initial 24
Molecules (1957-1968) and highly pathogenic avian influenza
June 1994
H5N1 strains within the Goose/Guangdong/96-like
H5 lineage (HPAI H5N1),” apart from in a few
select cases as detailed in Sections III-D-7-a and –b
where containment can be brought down to BL-2
(p.21-22). Both SARS -CoV and MERS-CoV are
classed as Risk Group 3 pathogens (high individual
risk, low community risk) on a 1 to 4 scale (with 4
being high individual risk and high community risk
pathogens).


? Functions (1=low
5=high)
CDC Public Health

Guidance for Supplement F of the CDC's 2005 guidance document
Community-Level on SARS-CoV provides laboratory biosafety
Preparedness and guidelines for working with specimens associated
3 May
Response to Severe Guidance Safety 1 with SARS-CoV. The document focuses on Link
2005
Acute Respiratory preparedness for a potential public health response
Syndrome (SARS- (diagnostics, specimens potentially containing
COV), Supplement F SARS-CoV)
"Laboratory Guidance"
42 CFR 73.12 notes that those designing a biosafety

plan "should consider" this document. NIH and
Biosafety in Amended
CDC-led "national code for biosafety." A thorough
Microbiological and December
Safety & 12345678 set of best practices for biosafety risk assessment and
Biomedical Guidance 2009; 5 Link
Security 90A implementation. Includes a chapter on biosecurity.
Laboratories (BMBL), initial
Chapters are broken down into separate entries in
5th edition 1984
this dataset due to the large scope and relevance of
the content.
Provides an overview of the principles of biosafety

and biosecurity, relevant to the rest of the BMBL
… "Section I - Safety & 12345678
5 guidance. While the substance is not in this section,
Introduction" Security 90A
it outlines the ways to think about biosafety and how
security fits into that context
Overview of the factors to consider and the methods

… "Section II -
Safety & 12345678 to use when determining the risk of work with a
Biological Risk 5
Security 90A given infectious agent. Includes 2 primary
Assessment"
categories: agent hazards and procedure hazards.


? Functions (1=low
5=high)
Overview of the principles to biosafety, factors

involved in choosing a BSL, import/shipment, select
… "Section III - Safety & agents. This chapter points to other chapters and
123456A 5
Principles of Biosafety" Security appendices for more detail. While not explicitly
about biosecurity, these concepts overlap with
security measures
Specifications for BSL 1-4 lab safety, including

… "Section IV -
Safety & standard microbiological practices, special practices,
Laboratory Biosafety 12346A 5
Security safety equipment (primary barriers/PPE), and
Level Criteria"
laboratory facilities requirements for each BSL level.
… "Section V - Specifications for ABSL 1-4 lab safety, including

Vertebrate Animal standard microbiological practices, special practices,
Biosafety Level 12346A 5 safety equipment (primary barriers/PPE), and
Criteria for Vivarium laboratory facilities requirements for each ABSL
Research Facilities" level.
… "Section VI - Considerations for planning and implementing a

Principles of Safety & biosecurity program (examples, not standards).
12345689 5
Laboratory Security Relationship between security measures and safety
Biosecurity" measures.
… "Section VII - Best practices for detection and mitigation of

Occupational Health laboratory-acquired infections (LAIs), may include
Safety 16 5
and medical exams, vaccines, reporting procedures,
Immunoprophylaxis" testing for exposed employees, and treatment plans


? Functions (1=low
5=high)
Background, laboratory and natural modes of

transmission, and laboratory safety and containment
recommendations for a large variety of pathogens
… "Section VIII -
Safety & and toxins. This are not comprehensive guidance but
Agent Summary 125689 5
Security starting points for safety and security planning.
Statements"
Recommends BSL levels and additional special
measures for research with many different
pathogens.
… "Appendix A -
Primary Containment
for Biohazards: Detailed information about how to set up a biosafety
Safety 16A 5
Selection, Installation cabinet
and Use of Biological
Safety Cabinets"
… "Appendix B -
Disinfection and sterilization procedures, planning,
Decontamination and Safety 16A 5
and characterization
Disinfection"
… "Appendix C -
Safety & Shipping, transport, and transfer codes and some
Transportation of 5 5
Security summary guidance
Infectious Substances"
… "Appendix D - Requirements for BSL-3-Ag, and BSL-3, Enhanced.

Safety &
Agriculture Pathogen 12346A 5 Requires BSL-3-Ag containment for all work with
Security
Biosafety" HPAI
… "Appendix E - References to the Arthropod Containment Levels

Arthropod Containment Safety 0 0 and guidelines developed by the American
Guidelines (ACG)" Committee of Medical Entomology


? Functions (1=low
5=high)
… "Appendix F -
Safety & 12345678 References applicable Select Agents and Toxins
Select Agents and 5
Security 9 codes, summarizes those codes.
Toxins"
… "Appendix G -
Integrated Pest Safety 2689 2 Pest control specifications and guidance
Management (IPM)"
… "Appendix H - Describes requirements and recommendations for

Working with Human, working with human/primate/mammalian cells,
NHP and Other Safety 1260A 1 including OSHA regulations, recommended
Mammalian Cells and prophylactic vaccinations, and recommended
Tissues" handling practices/risk assessments.
… "Appendix I - Recommendations for training, facilities planning,

Guidelines for Work Safety & safety equipment, handling aerosols/spills/sharps
124689 1
with Toxins of Security incidents, safety precautions and waste
Biological Origin" disposal/decontamination.
… "Appendix J - NIH
Oversight of Research Introduction to the NIH rDNA guidelines and the
Safety 0 5
Involving Recombinant IBC/RAC processes.
Biosafety Issues"
Public Health Service

and NIH Office of
Laboratory Animal Revised
Policy 0 Policy ensuring animal welfare Link
Welfare "Policy on 2015
Humane Care and Use
of Laboratory Animals"


? Functions (1=low
5=high)
International guidance on ensuring laboratory animal

International Guiding
welfare prepared by the Council for International
Principles for December
Guidance 0 Organization of Medical Sciences (CIMS) and the Link
Biomedical Research 2012
International Council for Laboratory Animal Science
Involving Animals
(ICLAS)
Guidance on ensuring laboratory animal welfare and

Guide for the Care and
ethical use of animals, prepared by the National
Use of Laboratory
Guidance 2011 0 Research Council of the National Academies' Link
Animals, Eighth
Committee for the Update of the Guide for the Care
Edition
and Use of Laboratory Animals.
ASHRAE laboratory
design guide, 1st Guidance Safety 3 Technical book on laboratory design Book
edition
A WHO publication that provides guidance on topics

including: laboratory biosafety, codes of practice in
laboratories, laboratory equipment operation, good
microbiology techniques, contingency and
emergency planning, disinfection and sterilization,
WHO Laboratory 2004;
transport of infectious substances, biosafety
Safety Manual, 3rd Guidance initial Safety 5 Link
considerations for recombinant DNA technology,
Edition 1984
hazardous chemicals, fire and electrical safety, the
concept of a biosafety officer and a biosafety
committee, ensuring the safety of support (repair,
cleaning) staff, and safety checklists. The manual
also has two pages on biosecurity.


? Functions (1=low
5=high)
WHO guidance for all member states on addressing

biosecurity issues. The proposed approach is to start
by identifying valuable (and/or particularly
WHO Biorisk
Septembe dangerous) biological material that needs to be
Management: Security 3 Link
r 2006 safeguarded. The document confirms that “there is
Biosecurity Guidance
no international agreement on what kind of biosafety
containment level and laboratory biosecurity
practices should apply for specific situations” (p.21).
ABSA biosecurity task

force white paper: January
Guidance Security 1 A one-page document on biosecurity Link
understanding 2003
biosecurity
The European Committee for Standardization (CEN)

convened a workshop, “CEN Workshop 31-
CEN Workshop
Laboratory biosafety and biosecurity,” which
Agreement, CWA Safety
February resulted in this agreement. The agreement covers
15793 “Laboratory Guidance and 1 Link
2008 both biosafety and biosecurity risks. The United
biorisk management Security
States is not a CEN member, but there was U.S.
standard”
participation at the workshop (both direct and
through a public comment process).
US Department of
This document helps practitioners comply with U.S.
Transportation,
Department of Transportation regulations on
"Transporting Guidance Safety 5 1 Link
hazardous material transportation (49 CFR 171 to
Infectious Substances
180).
Safely"


? Functions (1=low
5=high)
The Dangerous Goods Regulations, 56th edition, is

prepared by the International Air Transport
No
IATA Dangerous Association (IATA). It includes guidance for air
Guidance Safety 5 1 free
Goods Regulations transport of infectious substances, but the guidance
copy.
document is not freely available and must be
purchased.
A brief and now-out-of-date guidance document

regarding air transport of infectious substances.
Dedicates one page (p.5) to spill mitigation
procedures and first aid. The document is valid for
IATA Guidance Now out
the Dangerous Goods Regulations 52nd edition from
Document: Infectious Guidance of date; Safety 5 1 Link
2010, whilst the current regulations are on their 56th
Substances 2010
edition. IATA also offers a training course on the
topic, but it is not free
(http://www.iata.org/training/courses/Pages/infectiou
s-substances-tcgp43.aspx)
WHO has published guidance on regulations (i.e.

Applicabl model regulation) on the transport of infectious
e1 substances every two years, based on the broader
WHO Guidance on
January recommendations established as the "United Nations
regulations for the
Guidance 2015, Safety 5 1 Recommendations on the Transport of Dangerous Link
Transport of Infectious
covers Goods" program. The document includes specific
Substances 2015-2016
2015- guidance for topics such as: shipping medical waste
2016 and infected animals; transport by air, rail, road, sea,
and post; and spill emergency response regulations.


? Functions (1=low
5=high)
Strengthening Risk The document provides an overview of means to

Governance in reduce biosafety and biosecurity risks, and in
Safety
Bioscience December carrying out a thorough risk appraisal process.
Guidance and Link
Laboratories,” Sandia 2009 Appendix C describes the BioRAM Model, which is
Security
National Laboratories, an algorithm to appraise the risk at the pathogen-
SAND2009-8070 specific level.

16.15 Restriction of Fundamental Research, Dual Use Research of Concern and
Recombinant DNA Guidelines
The current U.S. Government’s deliberative process and pause of certain GoF research, relates to years of
discussion and policymaking for scientific research that could be used for beneficial or military/harmful
purposes. Federal policies on the dual use of scientific research encompass the export control regime,
communication of fundamental scientific research, recombinant DNA guidelines, and policies on
oversight of dual use life sciences research. Export control requirements are incorporated in the detailed
assessment of security measures in Appendix V: Sec 16.11. Policies on communication of fundamental
research, dual use life sciences research of concern, and recombinant DNA provide the overarching
framework under which past and future life science research occurs. Because they are central to
biosecurity considerations of GoF pathogens, but are not physical or personnel security measures, these
policies are briefly described below.
16.15.1 National Security Decision Directive 189
In 1982, President Reagan issued National Security Decision Directive (NSDD)-189, National Policy on
the Transfer of Scientific, Technology and Engineering Information,2812 which states:
• “that, to the maximum extent possible, the product of fundamental research remain unrestricted,”
• “that, where the national security requires control, the mechanism for control of information
generated during federally-funded fundamental research in science, technology and engineering
at colleges, universities and laboratories is classification,” and
• that “no restrictions may be placed upon the conduct or reporting of federally-funded
fundamental research that has not received national security classification, except as provided in
applicable U.S. statutes.”
In this policy, fundamental research is defined as “basic and applied research in science and engineering,
the results of which ordinarily are published and shared broadly within the scientific community, as
distinguished from proprietary research and from industrial development, design, production, and product
utilization, the results of which ordinarily are restricted for proprietary or national security reasons.” This
policy repeatedly has been upheld since its issuance.
16.15.2 Dual Use Life Sciences Research Concern
The 2012 debate about publication of specific mutations in the H5 influenza gene that resulted in
mammalian transmissible H5 influenza viruses catalyzed the issuance of the United States Government
Policy for Oversight of Life Sciences Dual Use Research of Concern2813 in March 2012 and the United
States Government Policy for Institutional Oversight of Life Sciences Dual Use Research of Concern in
September 2014.2814 These policies established requirements for review and oversight of life sciences
2812 President Ronald Reagan. National Security Decision Directive 189 – National Policy on the Transfer of Scientific,
Technical and Engineering Information. September 21, 1985.
2813 United States Government Policy for Oversight of Life Sciences Dual Use Research of Concern. (2012) Accessible at
http://www.phe.gov/s3/dualuse/documents/us-policy-durc-032812.pdf. Accessed on September 9, 2015.

research involving one of 15 agents (14 pathogens and one toxin), which includes highly pathogenic avian
influenza virus and the reconstructed 1918 influenza virus, and one of seven categories of experiments
that raise particular concern:
1. Enhancement of harmful consequences of certain agents or toxins,
2. Disruption of immunity or the effectiveness of an immunization against certain agents or toxins

without clinical or agricultural justification,
3. Alteration of an agent or toxin to confer resistance to clinically or agriculturally useful

prophylactic or therapeutic interventions against the agent or toxin or to facilitate evasion of
detection methodologies,
4. Increase in the stability, transmissibility, or ability to disseminate certain agents or toxins,
5. Alteration of the host range or tropism of certain agents or toxins,
6. Enhancement of susceptibility of a host population to certain agent or toxins, and
7. Generation or reconstitution of an eradicated or extinct agent or toxin.
U.S. government agencies that fund life sciences research are required to develop agency-specific
requirements to implement the Federal (March 2012) “dual use research of concern” (DURC) policy. The
September 2014 institutional DURC oversight policy describes an organizational framework for oversight
of research that has dual use potential and provides a list of responsibilities for the institution, principal
investigator, and federal government. The National Institutes of Health provides a Companion Guide for
the dual use policies, which includes identification and assessment of research, a framework for
institutional review, development and review of a risk mitigation plan, and communication of research
with dual use potential.2815 In addition, the NIH provides a series of case studies to assist scientific
organizations implement the policy.2816 These case studies are intended to illustrate how to apply the
policy to the review of life science research.
In practice, the Institutional Biosafety Committees of several academic and nonprofit research institutions
review research for dual use potential.2817 Some institutions have established specific committees who
review research for its dual use potential.2818 In 2012, some research institutions stopped reviewing
research for its dual use potential because they no longer conduct select agent research.2819 However,
some of these institutions may resume reviewing research for dual use potential if they conduct research
with any quantity of botulinum toxin, per the September 2014 institutional DURC oversight policy.2820
2815 National Institutes of Health. Tools for Identification, Assessment, Management, and Responsible Communication of Dual
Use Research of Concern. A Companion Guide to the United States Government Policies for Oversight of Life Sciences
Dual Use Research of Concern. Sept 2014. Accessible at http://www.phe.gov/s3/dualuse/Documents/durc-companion-
guide.pdf. Accessed on September 18, 2015.
2816 National Institutes of Health. Implementation of the USG Policy for Institutional Oversight of Life Sciences DURC:
Illustrative Case Studies. September 2014. Accessible at http://www.phe.gov/s3/dualuse/Documents/12-case-studies-
durc.pdf. Accessed on September 18, 2015.
2817 AAAS, AAU, APLU, FBI. Bridging Science and Security for Biological Research: A Discussion about Dual Use Review
and Oversight at Research Institutions. Workshop Report. 2012. Accessible at http://www.aaas.org/report/discussion-about-
dual-use-review-and-oversight-research-institutions. Accessed on September 18, 2015.
2818 Ibid.
2819 Ibid.

16.15.2.1 HHS Framework for H5N1 and H7N9
In 2013, the U.S. Department of Health and Human Services (HHS) issued its framework for guiding
funding decisions on H5N1 and H7N9 GoF research, specifically that which involves transmission among
mammals by respiratory droplets.2821,2822,2823
The H5N1/H7N9 framework builds on the existing funding agency “standard review” process for GoF
research that increases aerosol transmission of the viruses. The standard review process involves an initial
peer review for scientific merit and subsequent dual use review if the research meets the U.S. government
definition for “dual-use research of concern,” as stipulated by the March 2012 DURC policy. 2824,2825,2826
Once this standard review process has been completed, projects “reasonably anticipated to generate an
HPAI H5N1 virus that is transmissible between mammals by respiratory droplets” must meet the
following seven criteria before it can be considered for funding by an HHS funding entity:
1. The resultant virus “could be produced through a natural evolutionary process,”
2. The project would address “a scientific question with high significance to public health,”
3. “No feasible alternative methods [exist] to address the same scientific question in a manner that
poses less risk” than the proposed project,
4. The potential biosafety risks “to laboratory workers and the public can be sufficiently mitigated
and managed,”
5. The biosecurity risks “can be sufficiently mitigated and managed,”
6. The research is “anticipated to be broadly shared in order to realize its potential benefits to global
health,” and
7. The research “will be supported through funding mechanisms that facilitate appropriate oversight
of the conduct and communication of the research.” 2827
If a project meets all seven criteria as determined by the HHS funding entity, it enters a HHS department-
level review to determine whether the proposal is acceptable for HHS funding based on the following
considerations:2828 A) the quality of the risk assessments; B) additional factors that may affect the
decision; C) required risk mitigation measures; and D) the project’s place within the broader HHS
2821 A Framework for Guiding U.S. Department of Health and Human Services Funding Decisions about Research Proposals
with the Potential for Generating Highly Pathogenic Avian Influenza H5N1 Viruses that are Transmissible among Mammals
by Respiratory Droplets, p. 4, https://www.phe.gov/s3/dualuse/Documents/funding-hpai-h5n1.pdf.
2822 Harold Jaffe, Amy P. Patterson, Nicole Lurie, “Extra Oversight for H7N9 Experiments,” appeared in
Science (Letters) 341, no. 6147 (7 August 2013): p.713-714, http://www.sciencemag.org/content/341/6147/713.2.full.
2823 Harold Jaffe, Amy P. Patterson, Nicole Lurie, “Extra Oversight for H7N9 Experiments,” Nature (Correspondence) 500, no.
151 (8 August 2013), http://www.nature.com/nature/journal/v500/n7461/full/500151a.html.
2824 Amy P. Patterson et al., “A Framework for Decisions About Research with HPAI H5N1 Viruses,” Science (Policy Forum)
339, no. 6123 (1 March 2013): p. 1037, http://www.sciencemag.org/content/339/6123/1036.
2825 United States Government Policy for Oversight of Life Sciences Dual Use Research of Concern, p.1-2,
http://www.phe.gov/s3/dualuse/Documents/us-policy-durc-032812.pdf.
2826 United States Government Policy for Oversight of Life Sciences Dual Use Research of Concern. (2012) Accessible at
http://www.phe.gov/s3/dualuse/documents/us-policy-durc-032812.pdf. Accessed on September 9, 2015.
2827 A Framework for Guiding U.S. Department of Health and Human Services Funding Decisions about Research Proposals
with the Potential for Generating Highly Pathogenic Avian Influenza H5N1 Viruses that are Transmissible among Mammals
by Respiratory Droplets, p. 4.
2828 Ibid.

H5N1/H7N9 influenza portfolio.2829 If the departmental review results in a positive determination, the
project may be funded.2830 For all HHS-funded H5N1/H7N9 projects, researchers must report to HHS
“any unanticipated results that involve the generation of a virus that is transmissible among mammals by
respiratory droplets.”2831
16.15.3 NIH Guidelines for Research Involving Recombinant DNA and Synthetic Nucleic Acid
Molecules
The NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules describes
safety practices and containment procedures for life sciences research involving recombinant or synthetic
nucleic acid molecules. Synthetic nucleic acid molecules were added to the Guidelines in 2013.2832 The
purpose of the Guidelines is “to specify the practices for constructing and handing: (i) recombinant
nucleic acid molecules, (ii) synthetic nucleic acid molecules, including those that are chemically or
otherwise modified but can base pair with naturally occurring nucleic acid molecules, and (iii) cells,
organisms, and viruses containing such molecules.”2833 The NIH Guidelines are contractually required for
any institution receiving support from the National Institutes of Health. Other U.S. government agencies
also require compliance with the NIH Guidelines for receipt of life science grants involving recombinant
DNA.2834,2835
The NIH Guidelines require research institutions to assess and categorize the risk of research involving
recombinant or synthetic nucleic acid molecules by Risk Group, which are defined in the Guidelines.2836
They provide details about:
• The level of containment of research based on the experiments involved to prevent environmental
release of microorganisms, plants, or animals that contain recombinant or synthetic nucleic acid
molecules,
• Requirements for Institutional Biosafety Committees (IBCs) to review, approval and oversight of
research involving recombinant or synthetic nucleic acid molecules,
• Composition of the IBC,
• Experiments covered by the NIH Guidelines, including

2829 Ibid.
2830 Ibid.
2831 Ibid.
2832 National Institutes of Health. Frequently Asked Questions: NIH Guidelines for Research Involving Recombinant or
Synthetic Nucleic Acid Molecules. Accessible at http://osp.od.nih.gov/sites/default/files/Synthetic_FAQs_April_2013.pdf.
Accessed on September 18, 2015.
2833 National Institutes of Health. NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules
(NIH Guidelines). Nov 2013. Accessible at http://osp.od.nih.gov/sites/default/files/NIH_Guidelines.html#_Toc351276217.
2834 Department of Defense. General Guidelines for Awards Funded by the Department of Defense (DoD). Accessible at
http://www.usamraa.army.mil/pages/pdf/General_Guidelines_for_Awards_Funded_by_the_DoD.pdf. Accessed on
September 18, 2015.
2835 National Institutes of Health. Frequently Asked Questions: NIH Guidelines for Research Involving Recombinant or
Synthetic Nucleic Acid Molecules. Accessible at http://osp.od.nih.gov/sites/default/files/Synthetic_FAQs_April_2013.pdf.
Although no reference is includes, other U.S. Departments and Agencies, such as the Department of Homeland Security,
includes compliance with the NIH Guidelines as a requirement for receiving research funding.
2836 National Institutes of Health. NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules
(NIH Guidelines). Nov 2013. Accessible at http://osp.od.nih.gov/sites/default/files/NIH_Guidelines.html#_Toc351276217.

o Experiments that must be reviewed by IBCs, such as experiments involving transfer of
drug resistance traits into microorganisms and cloning of genes for toxin molecules,
o Experiments that require IBC and Institutional Review Board (for human subjects
research) approval,
o Experiments that require IBC approval before initiation for each Risk Group,
o Experiments involving infectious DNA or RNA viruses or defective DNA and RNA
viruses in the presence of helper viruses,
o Experiments involving animals or plants, and
o Experiments involving influenza viruses,
• Experiments that are exempt from the NIH Guidelines,
• Roles and Responsibilities of the research institution, principal investigator, and NIH,
• Information to be submitted to the NIH,
• Major and minor actions, and
• Responsibilities and composition of the Recombinant DNA Advisory Committee (RAC), who
provide advice on matters concerning research that involves recombinant or synthetic nucleic
acids, and may review and approve certain experiments.
The NIH provides information about major actions taken and experiments that are exempt from the NIH
Guidelines.2837
All academic, non-profit, and for-profit research institutions receiving NIH or other U.S. government
research funding are required to have an IBC that is registered with the NIH. Although the guidance is
voluntary, several companies that do not receive federal funding have also established an IBC that is
registered with the NIH.
16.16 Analysis of Security Measures: Requirements, Implementation, and Gaps of

Security Measures
Security measures reviewed for this assessment fall into seven categories: training; personnel reliability;
physical security; surveillance and monitoring; storage, inventory, and accountability processes; transfer,
shipment, and chain-of-custody protocols; and emergency response. In this section, we define each of
these categories and describe requirements and implementation practices in non-Select Agent, Select
Agent, and Tier 1 Select Agent operating environments.
2837 National Institutes of Health. NIH Guidelines website. Accessible at http://osp.od.nih.gov/office-biotechnology-

activities/biosafety/nih-guidelines. Accessed on September 18, 2015.

16.16.1 Training
Training is required to ensure that all workers understand the risks associated with their research, and
appropriate measures to address those risks.
16.16.1.1 General – At all levels
Requirements
General U.S. labor laws enforced by OSHA require all biological researchers to received basic safety
training. In addition, OSHA requires employees and students working in research laboratories receive
training on exposure to hazardous chemicals, hazard communication, bloodborne pathogens, personal
protective equipment, eye and face protection, hand protection, and respiratory protection.2838
The Biosafety in Microbiological and Biomedical Laboratories describe training associated with different
biosafety level laboratories.2839 The primary goal of training personnel in appropriate laboratory safety
procedures, including preventing, detecting, and reporting unsafe behavior, is to reduce the risk of
accidental exposure. Typical biosafety training includes practical procedures for working in the
laboratory, donning required personal protective equipment, signage indicating the hazards in the
laboratory and emergency contacts, reporting procedures in case of laboratory accidents or negligence,
and decontamination measures in case of an accident. In addition, scientists should receive training to
demonstrate technical proficiency at the appropriate biosafety level and to demonstrate knowledge about
hazards of specific infectious agents.
In addition, scientists conducting federally-funded research with any quantity of botulinum toxin that
involves at least one of the seven categories of experiments, and is considered to have dual use potential
are required to receive DURC training by their institutions.2840
Implementation at Research Institutions
Research institutions provide training to employees and staff on basic laboratory safety, materials safety,
bloodborne pathogens, hazard waste disposal, use of sharps, and information security. Several institutions
provide training on dual use research of concern and recombinant DNA guidelines.
Researchers working in high containment laboratories receive agent-specific training, which includes
understanding clinical symptoms, several weeks of hands-on mentored training, and knowledge of
standard operating procedures. Many laboratories conduct hands-on, mentored training in stages, allowing
new recruits to proceed to higher containment levels (BSL-3) only after they have demonstrated
competence at a lower containment level (BSL-2) or same containment level (BSL-3). This accompanied
training process ensures that each individual demonstrates proficiency and competency in conducting
experiments safely and according to standard operating procedures. In addition, researchers are informed
of facility security and access, visitor access, entry requirements, and facility-specific policies in addition
to training about waste management, emergency response, and use of sharps in high containment.2841
2838 Occupational Safety and Health Administration. Laboratory Safety Guidance. OSHA 3404-11R 2011. Accessible at
https://www.osha.gov/Publications/laboratory/OSHA3404laboratory-safety-guidance.pdf. Accessed on September 18, 2015.
2839 Centers for Disease Control and Prevention and National Institutes of Health. Biosafety in Microbiological and Biomedical
Laboratories, 5th Edition. 2009. Accessible at http://www.cdc.gov/biosafety/publications/bmbl5/BMBL.pdf. Accessed on
September 18, 2015.
2841 Lesley C. Homer et al., “Guidelines for Biosafety Training Programs for Workers Assigned to BSL-3 Research
Laboratories,” Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science 11, no. 1 (2013): p.13.

Laboratories working with animals tend to have a heightened awareness of the potential for animal rights
extremists trying to gain access to research facilities or recruiting scientists to assist their efforts.
16.16.1.2 Additional measures at the Select Agent level
Requirements
In addition to the training requirements for all laboratories and high-containment laboratories, the BSAT
regulations specifically require employee security awareness training. Each entity that is registered to
possess Select Agents must have a security plan.2842 The security plan must have provisions to ensure
“that all individuals with access approval [to a Select Agent] understand and comply with the security
procedures” described in the plan.2843 The entity must implement these measures by providing
information and training on security topics, such as security awareness, to any individual with approval
and access to BSAT facilities.2844 Training for employees with Select Agent access must be conducted at
least once a year and a written record must be kept that details the training, including “the means used to
verify that the employee understood the training.”2845
In addition, scientists conducting federally-funded research with at least one of the 15 agents that involves
at least one of the seven categories of experiments and is considered to have dual use potential are
required to receive DURC training by their institutions.2846
Researchers approved to work with BSAT receive hands-on, mentored training similar to that described
for research in high containment laboratories. Some of the BSAT laboratory personnel interviewed
described training routines that went well beyond the minimum required to meet regulatory requirements.
Trainers and mentors test the knowledge gained by researchers to assess the degree to which they
understand the standard operating procedures and laboratory safety and security practices.
Research institutions provide training to BSAT approved staff about security considerations associated
with working in BSAT laboratories. One institution offers insider threat training provided for Tier 1
BSAT researchers to non-Tier 1 BSAT researchers; the non-Tier 1 BSAT researchers also attend this
training. Another institution provides training on security risks and updates this training as information on
threats presents itself.
Institutions train scientists through laboratory drills and exercises, which is described in the Emergency
Response section. Research institutions are encouraged to train individuals on defining and responding to
“suspicious activity” and appropriate responses to security emergencies.2847
2842 U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security,” <http://www.ecfr.gov/cgi-bin/text-
idx?SID=94bd3a730b8387eb15bc058bc4637627&mc=true&node=se42.1.73_111&rgn=div8>.
2843 U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2844 42 CFR 73.15. U.S. Government Publishing Office, “Title 42: Public Health, §73.15 Training,” <http://www.ecfr.gov/cgi-
bin/retrieveECFR?gp=&SID=0f9c35f1c983d1e04a020889c033b02b&mc=true&n=pt42.1.73&r=PART&ty=HTML#se42.1.
73_115>.
2845 42 CFR 73.15. U.S. Government Publishing Office, “Title 42: Public Health, §73.15 Training.”
2847 LouAnn C. Burnett, “Biosafety Practices Associated with Potential Agents of Biocrime and Biowarfare,” Emerging
Technologies, Supplement 3, 1A.2.5.

16.16.1.3 Additional measures at the Tier 1 Select Agent level
Requirements
Institutions are required to provide insider threat training to Tier 1 BSAT researchers. Because insider
threat trainings are only required for personnel working with Tier 1 BSAT, it not mandatory for the
pathogens considered in this report under current regulations.2848
Several institutions that support Tier 1 BSAT research provide insider threat training to appropriate
researchers. One institution has their local FBI Weapons of Mass Destruction Coordinator conduct the
training. This institution offers the training to its non-Tier BSAT researchers, several of whom attend
voluntarily. Other institutions provide their own insider threat training.
16.16.1.4 Gap analysis
Based on the above information, the following gaps were identified:
• The ability of training provided to inculcate security awareness at non-BSAT facilities,

particularly at facilities that do not work with animals, is unclear. Security recommendations
provided in guidance documents at the non-BSAT level contain little guidance to enhance
security awareness among employees and staff. 2849 Current text in authoritative guidance
documents, including first and foremost the BMBL, is written to assist laboratory managers in
implementing a security plan, but provides little to no advice for the average laboratory workers
to become more security-conscious.2850 and
• At all levels, the ability to maintain high-quality training depends on laboratory personnel
resources and funding. BSAT approved research institutions do not receive additional financial
support to pay for additional staff dedicated to training BSAT researchers.
16.16.2 Personnel Reliability
Personnel reliability measures seek to prevent insider threats through initial vetting and periodic
monitoring of employees and students with access to BSAT. This vetting process involves both
background checks and related measures, and reliability assessments, which include demonstration of
competency and proficiency in high containment laboratories. Similar demonstrations of competency and
proficiency in high containment laboratories are conducted in laboratories that are not regulated by the
BSAT Regulations.2851
2848 This is specified under 73.15b). U.S. Government Publishing Office, “Title 42: Public Health, 73.15 Training,”
<http://www.ecfr.gov/cgi-
bin/retrieveECFR?gp=&SID=0f9c35f1c983d1e04a020889c033b02b&mc=true&n=pt42.1.73&r=PART&ty=HTML#se42.1.
73_115>.
2849 In a 2003-2004 survey of Select Agent researchers conducted by Sandia National Laboratories, 53% of respondents state
that their facilities provide biology-specific security training, versus 26.5% who said they did not. 73.5% of respondents
further stated that the security training was done in conjunction with biosafety training. Sandia National Laboratories,
“Laboratory Biosecurity: A Survey of the U.S. Bioscience Community,” SAND No. 2006-1197P, Unlimited Release,
February 2006, p. 6, http://www.biosecurity.sandia.gov/ibtr/subpages/pdfs/surveyResponses022606.pdf.
p.104-113.
2851 AAAS. Biological Safety Training Programs as a Component of Personnel Reliability. Workshop Report. 2009. Accessible
at http://www.aaas.org/sites/default/files/AAAS-Biosafety-report.pdf. Accessed on September 18, 2015.

16.16.2.1 General– At all levels
Requirements
In non-security intensive environments, personnel reliability measures serve three purposes: 1) to ensure
that only personnel who are able to work competently and reliably under high containment conductions
have access to the laboratories; 2) to identify and address potential issues that may increase an
individual’s propensity to make mistakes or act negligently in the laboratory; and 3) identify and
appropriately address export control requirements.
The description about competency and proficiency training in the previous section inform personnel
reliability measures in high containment laboratories.2852
Deemed exports refers to the release of technology subject to the Export Administration Requirements
(EAR) for biological research to a foreign national who is not a permanent resident or protected
individual.2853 This Export Administration Regulation (EAR) applies to pathogens restricted by the
Australia Group and Select Agents.2854 All research in the United States is subject to EAR, except if the
technology is part of fundamental research, publicly available, or has been or will be published among
other exceptions. With respect to biological research, most research conducted at university laboratories
are not subject to EAR because it is considered fundamental research (the definition of which is the same
as in NSDD-189). If the research is not considered fundamental because restrictions have been applied
(e.g., restrictions on publication and proprietary information), it is subject to deemed export regulations. If
the research involves controlled pathogens, a determination of the conductions for information sharing
with a foreign national must be undertaken. Thus, the need for deemed export licenses appears to depend
not on the pathogen per se, but the conditions associated with the research, such as restricted publication.
In addition, the International Traffic in Arms Regulations require foreign nationals would need a permit to
access any biological agents “modified to increase…capability to produce casualties in humans or
livestock.”2855
As described in the previous Training section, several research institutions provide hands-on, mentored
training to researchers to ensure they demonstrate competency and proficiency of standard operating
procedures and biosafety. In addition, several research institutions promote an opt-out policy to encourage
researchers to voluntarily remove themselves from the laboratory if they are experiencing personal issues,
such as illness, exhaustion, or personal distractions.2856 These researchers are not punished; they are able
to return to laboratory work once the distraction has been resolved.
2852 Ibid.
2853 Department of Commerce. Deemed Exports and Fundamental Research for Biological Items. Accessible at
https://www.bis.doc.gov/index.php/policy-guidance/product-guidance/chemical-and-biological-controls/14-policy-
guidance/deemed-exports/111-deemed-export-and-fundamental-research-for-biological-items. Accessed on September 18,
2015.
2854 U.S. Department of Commerce, Commerce Control List, “Category 1 – Special Materials and Related Equipment,
Chemicals, ‘Microorganisms’ and ‘Toxins’,” <http://www.bis.doc.gov/index.php/forms-documents/doc_download/989-
ccl1>.
2855 22 CFR 121.1(XIV)(b) U.S. Government Publishing Office, “The United States Munitions List,” <http://www.ecfr.gov/cgi-
bin/text-idx?SID=88e7fab9254a3319c3df6fb11a2233ab&mc=true&node=se22.1.121_11&rgn=div8>.
2856 AAAS. Biological Safety Training Programs as a Component of Personnel Reliability. Workshop Report. 2009. Accessible
at http://www.aaas.org/sites/default/files/AAAS-Biosafety-report.pdf. Accessed on September 18, 2015.

Research institutions have dedicated offices to comply with export control requirements.2857 In addition,
some larger institutions designate an “Export Controls Coordinator” to help laboratories comply with
“deemed export” regulations.2858
Requirements
The BSAT Regulations require Security Risk Assessments (SRA) for all individuals seeking access to
BSAT.2859 SRA are required before initial approval and every three years. The assessment focuses on
denying access to individuals known or suspected of having committed a serious crime, use illegal drugs,
adjudicated as a mental defective, are a national of a country or acts on behalf of a country that has
“repeatedly provided support for acts of international terrorism” as determined by the Secretary of State,
or are themselves involved with terrorists or organized criminals. The exact criteria can be found in the
HHS Select Agents regulations, more specifically under Title 42 “Public Health,” Code of Federal
Regulations (CFR) 73.10 “Restricting access to Select Agents and Toxins; security risk assessments.”2860
SRAs are conducted by the FBI Criminal Justice Information Services, who conduct a series of database
checks to assess whether applicants should be granted access to BSAT laboratories.2861
The BSAT Regulations require all individuals with access to Select Agents and Toxins to report
suspicious behavior or signs or evidence of a physical security or inventory accounting compromise,
which then enables the responsible official to respond and revoke access as necessary.2862 The regulations
also require that the laboratory have a reporting process in place so that employees know how to report
suspicious activity, and so that the responsible official knows how to pass on reports to the appropriate
law enforcement agencies as necessary.2863
The institution’s Responsible Official can suspend or revoke an individual’s access to Select Agents and
Toxins if necessary.2864
In addition to the SRA required by the BSAT Regulations, Army Regulation 50-1 details requirements for
biosurety that apply to all individuals who work with DoD materials.2865
2857 A simple internet search identifies institutional information about export control regulations at a number of research
institutions.
2858 For example: http://www.colorado.edu/vcr/export-controls/guidance/biological-agents
2859 U.S. Government Publishing Office, “Title 42: Public Health, §73.10 Restricting Access to select agents and toxins; security
risk assessments,” <http://www.ecfr.gov/cgi-bin/text-
idx?SID=5e7f78178a77b6cce99612612ade5aa4&mc=true&node=se42.1.73_110&rgn=div8>.
2860 U.S. Government Publishing Office, “Title 42: Public Health, §73.10 Restricting Access to select agents and toxins; security
risk assessments.”
2861 The candidate provides fingerprints and a completed FD-961 form. NSABB, “Guidance for Enhancing Personnel Reliability
and Strengthening the Culture of Responsibility: A Report of the National Science Advisory Board for Biosecurity,” p. 17.
2862 As per the reporting requirements under 42 CFR 73.11(7)(i)-(v). U.S. Government Publishing Office, “Title 42: Public
Health, §73.11 Security.”
2863 42 CFR 74.11(6)-(8). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2864 NSABB, “Guidance for Enhancing Personnel Reliability and Strengthening the Culture of Responsibility: A Report of the
National Science Advisory Board for Biosecurity,” p. 17.
2865 Headquarters of the Department of the Army, “Army Regulation 50-1: Biological Surety,” Nuclear and Chemical Weapons
and Materiel, Unclassified, p. 10-20, <http://www.apd.army.mil/pdffiles/r50_1.pdf>.

Despite the costs involved, several institutions conduct background checks of all employees as part of the
hiring practice. A few institutions are required to have personnel undergo several different types of
national and international personnel security evaluations.
Institutions that support BSAT research require individuals seeking access to the BSAT to undergo the
SRA process. Although these assessments occur every three years, the FBI reportedly performs additional
spot checks by running names through up-to-date databases roughly every six months.2866 The benefit of
these checks relies on the types of databases used and the information contained therein.
Though not prevalent, a few research institutions conduct terrorism database checks, fingerprint
employees, conduct a health assessment, and/or check international databases. In addition, some
institutions have implemented an employee tracking system to determine personnel actions and who has
or does not have access to facilities.
Good interactions with co-workers, support staff, administrators, and supervisors enhance personnel
reliability measures, a comment echoed in the specialized literature as a requirement for an effective
behavioral monitoring program.2867
Interviewees described the community of BSAT researchers, including those working with influenza and
SARS-CoV, as close-knit. This environment promotes observation and timely reporting of behavior
considered out-of-place or abnormal in the laboratory work space.
The self- and peer-reporting approach reduces insider threat risk and complements the required individual
security risk assessment, although its effectiveness greatly relies on the reporting and security culture of a
laboratory. The self- and peer-reporting approach will be ineffective in laboratories where workers and
managers fear retaliation, wish to avoid additional paperwork, have overwhelming trust in their
coworkers, or distrust their superiors. The effectiveness of the self- and peer-reporting approach also
depends on the laboratory’s leadership maintaining a close relationship with workers: “a leader who is
engaged with his or her staff, who greets them by name and is perceived as accessible and caring, is more
likely to be able to prevent an employee from becoming disgruntled, be aware of potential problems, and
be better able to intervene to prevent the employee from becoming a crisis.”2868
Requirements
Personnel reliability measures for Tier 1 Select Agents involve a pre-access suitability assessment and a
formal continuous suitability assessment process, in addition to the reliability measures established for
National Science Advisory Board for Biosecurity,” p.17-18;
National Science Advisory Board for Biosecurity, “Enhancing Personnel Reliability among Individuals with Access to
Select Agents,” Report for the National Science Advisory Board for Biosecurity, May 2009, p. 12,
<http://osp.od.nih.gov/sites/default/files/resources/NSABB%20Final%20Report%20on%20PR%205-29-09.pdf>.
As per 42 CFR 73.10(2)(j). U.S. Government Publishing Office, “Title 42: Public Health, §73.10 Restricting Access to
select agents and toxins; security risk assessments.”
2867 Ibid;
David R. Franz, Balancing Our Approach to the Insider Threat,” Biosecurity and Bioterrorism: Biodefense Strategy,
Practice, and Science 9, no. 3 (2011): p.206.
2868 David R. Franz, Balancing Our Approach to the Insider Threat,” Biosecurity and Bioterrorism: Biodefense Strategy,
Practice, and Science 9, no. 3 (2011), p.206.

Select Agents.2869 Personnel reliability reporting requirements are stricter at the Tier 1 level. Regulations
mandate the “self- and peer-reporting of incidents or conditions that could affect an individual’s ability”
to access, work with, or safeguard Select Agents.2870
Several institutions have established behavioral threat assessment teams to conduct the suitability
assessment of researchers seeking or approved to work with BSAT.2871 Other institutions have established
occupational health programs where appropriately trained staff conduct periodic behavioral assessments
of BSAT researchers.
• Despite the requirement and implementation of personnel security efforts, a self-radicalized

individual who has no criminal history and is careful not to communicate with extremists or other
criminals would be extremely difficult to detect and, hence, unlikely to be screened out.
Similarly, an individual with a pattern of threatening activities not reported to police, such as a
propensity of becoming easily angered and agitated, will not be flagged by the SRA.2872
Moreover, an insider has time to become radicalized, affiliated with a criminal organization,
dependent on illegal drugs, or otherwise vulnerable or malicious in between personnel reliability
checks.2873 and
• Currently, institutions do not have a single system in which information about BSAT approved
individuals can be stored and accessed by both police and research administers. Such a system
would allow administrators to highlight potentially issues and police to determine whether any
approved individual has gotten in trouble by the police.
16.16.3 Physical Security
Physical security measures are designed to prevent unauthorized access to the laboratory, in particular to
protect pathogens and research animals. Examples of physical security measures include locks, physical
barriers, security guards, restricted access policies, and a security guard.
Requirements
Minimal access control measures are defined by the biosafety requirements (see Figure 16.4 above).
Visitors to a laboratory at any level beyond BSL-1 must meet entry and exit requirements set by the
facility managers. A lab at BSL-2 must have self-closing lockable doors, and a lab at BSL-3 and above
must restrict access to the facility (i.e. through locked doors). U.S. laboratory design standards
2869 42 CFR 73.11 (f)(1),(3)(iii). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2870 42 CFR 73.11 (f)(3)(i). Ibid.
National Science Advisory Board for Biosecurity,” p.17-18.
2873 Initially, these checks were conducted once every five years, but these concerns led to the current 3-year regulation. See:
NSABB, “Guidance for Enhancing Personnel Reliability and Strengthening the Culture of Responsibility: A Report of the
National Science Advisory Board for Biosecurity,” p. 16.

incorporate certain unnamed required security components as per the NIH Design Requirements Manual
for Biomedical Laboratories and Animal Research Facilities.2874 All “NIH owned and leased new
buildings and renovated facilities” must comply with the NIH Design Requirements, and must therefore
include these unnamed security features.2875 These security features are detailed in a document, the NIH
Physical Security Design Requirements, and is not to be released to the public.2876
As summarized in Figure 16.4 above, certain physical security measures are required for facilities housing
animals. Research involving animals is conducted at an ABSL facility or at a BSL-3-Ag facility.
Vivarium security is emphasized in guidance and facility design documents for such facilities, including
the BMBL, in part as a result of the long history of incidents involving animal rights extremists. ABSL
standards recommend that such facilities have no windows. This precaution imposes a barrier to entry by
malicious actors by making them find out where the animals are stored and by preventing access through
breaking of windows.2877 If windows are nevertheless included in the facility design, the NIH Design
Requirements Manual for Biomedical Laboratories and Animal Research Facilities stipulates that
vivarium “windows must be designed to preclude the visualization of animals from outside of the
building and also to address security issues.”2878 All facilities housing animals must also have self-closing
doors. This feature defends against cases where a malicious actor would open animal cages in the hopes
of causing an animal release.
The door lock type is important, as different lock types present different access control and access
revocation benefits or challenges. Guidance in written documentation discourage the use of vulnerable
traditional locks with regular keys (lock-and-key systems), because of the ease with which such locks can
be picked, the necessity of physically retrieving keys from employees that are supposed to lose facility
access, the ease with which the keys can be duplicated, and the lack of personnel tracking functionality
given that all personnel keys are identical.2879 High security cores provide stronger protection without
introducing electronic vulnerabilities.2880 Card, code, and biometric locks are more secure than traditional
locks, and typically also have logging capability, enabling security personnel to verify who accessed the
laboratory at what time.2881 These electronic systems are favored in laboratories that have the funds to
incorporate them, as they facilitate billing users per hour of lab use and also double as a monitoring
feature. Whether any U.S. BSL-3 laboratories not working with Select Agents solely rely on weak lock-
and-key systems is a knowledge gap. Although discouraged by all authors, lock-and-key systems are still
listed as a security option in publicly-available literature on laboratory security design, mainly because it
remains the least costly access control system.2882
2874 The National Institutes of Health, Division of Technical Resources, “Design Requirements Manual,” p. 1-79,
<http://orf.od.nih.gov/PoliciesAndGuidelines/BiomedicalandAnimalResearchFacilitiesDesignPoliciesandGuidelines/Docum
ents/Design%20Requirements%20Manual/NIH%20Design%20Requirements%20Manual%20ver%205-13.pdf>.
2875 Ibid.
2876 Ibid.
2877 Ibid.
2878 Ibid.
2880 Ibid.
Chemical Hazards, Updated version, p. 257. Ibid, p. 257.
2882 Mechanical key locks remain a listed lock type noted in Appendix III: “Comparison of Access Control Devices and Systems
which are used to Control Access to Select Agents and Toxins”, of: Centers for Disease Control and Prevention (CDC)
Division of Select Agents and Toxins, Animal and Plant Health Inspection Service (APHIS) Agriculture Select Agent
Program, “Security Guidance for Select Agent or Toxin Facilities: 7 CFR Part 331, 9 CFR Part 121, 42 CFR Part 73,” p.33;
See also: Daniel D. Watch, Building Type Basics for Research Laboratories, second edition (Hoboken: John Wiley & Sons,
Inc., 2008), p.40.

In practice, BSL-3 laboratories have at least two physical barriers (and in many instances, several more
“layered defenses”) between an outsider and the laboratory space where pathogens are manipulated or
stored. Often, but not always, different types of access controls are used to allow access to laboratories.
These types of controls can be electronic, physical, or human.
The implementation of physical security measures at a facility has been reported in terms of the
approximate time that a hypothetical malicious actor with various hand-held breaching implements would
take to overcome the barrier.2883 In other words, the facility implements security measures that buy a
certain amount of time against malicious actor penetration. No such openly-available security standards
are available for labs that do not work with Tier 1 Select Agents. Moreover, openly-available regulations
do not stipulate what specific door lock types and door materials are to be employed or avoided for
physical barriers to secure a laboratory space.
Access policies make detection of unauthorized individuals easier. Many laboratories provide workers
with ID badges and restrict access to the laboratory after normal working hours unless night operations
are required for a research project or to provide animal care.2884 Identification complicates the task of a
malicious actor trying to pass as an authorized individual, and restricting operation times decreases the
chances that a malicious insider can carry out an act when no one else is around.
Most if not all high containment laboratories have special policies in place to restrict and control visitor
access, which in practice often revolve around ensuring that visitors are positively identified in some
manner and are escorted on site.2885
16.16.3.2 Additional measure at the Select Agent level
Requirements
BSAT Regulations require that physical security procedures be incorporated into the facility’s security
plan.2886 BSAT and animals exposed or infected with a BSAT must be access controlled and secured
“against unauthorized access, theft, loss, or release,” although the regulations do not detail how this must
be done.2887 “Freezers, refrigerators, cabinets, and other containers where select agents or toxins are
stored” must be “secured against unauthorized access,” and card access systems and lock boxes are
acceptable ways of doing so.2888
2883 For use as part of the Select Agent security planning process, see: Centers for Disease Control and Prevention (CDC)
Division of Select Agents and Toxins, Animal and Plant Health Inspection Service (APHIS) Agriculture Select Agent
Program, “Security Guidance for Select Agent or Toxin Facilities: 7 CFR Part 331, 9 CFR Part 121, 42 CFR Part 73,” p.30-
31, 42-43.
For general use, see for example: Betty E, Biringer, Rudolph V. Matalucci, Sharon L. O’Connor, Security Risk Assessment
and Management: A Professional Practice Guide for Protecting Buildings and Infrastructures (Hoboken: John Wiley &
Sons, Inc., 2007), p. 327.
2884 Ibid, p. 13;
National Research Council of the National Academies, Prudent Practices in the Laboratory: Handling and Management of
Chemical Hazards, Updated version, p. 258.
2885 LouAnn C. Burnett, “Biosafety Practices Associated with Potential Agents of Biocrime and Biowarfare,” Emerging
Technologies, Supplement 3, 1A.2.4;
Sandia National Laboratories, “Laboratory Biosecurity: A Survey of the U.S. Bioscience Community,” p.13.
2886 42 CFR 73.11(c)(1). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2887 42 CFR 73.11(c)(2) and 42 CFR 73.11(c)(8). U.S. Government Publishing Office, “Title 42: Public Health, §73.11
Security.”
2888 42 CFR 73.11 (c)(1)-(3). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”

The security plan and security training regulations also require measures that ensure a timely and
coordinated security response. The facility’s security plan must include procedures for “removing
unauthorized or suspicious persons,” immediately reporting suspicious persons and activities and
potential signs of inventory compromise, reporting potential criminal activity to the authorities, and
addressing access control compromises (such as lost keys) and access control revocation.2889
Several research institutions employee at least three barriers to prevent physical access to the BSAT
laboratory. These barriers are controlled using different types of locks to prevent anyone from access if
one unlocking mechanism is stolen. These different types can be physical, electronic, human, or
physiological.
A certain tradeoff exists between facility security measures and making the facility hard-to-find for
external malicious actors, although both measures help ensure physical security of the facility. Employing
measures, such as fences, around an institution may enhance physical security, but also draws attention to
the facility and singles it out for malicious actors. Some institutions have taken measures to not call
attention to buildings wherein BSAT research is conducted to prevent targeting by malicious actors.
Institutions routinely review laboratory access records to identify any anomalies in laboratory access.
The effectiveness of any physical control system is only as good as the responsiveness of the employer to
revoke access to ex-employees, particularly in cases where the individual may become malicious as a
result of their termination. Modern electronic access control systems often enable a designated security
official to disable an individual’s access at any time. This security feature is implemented at some
campuses, where campus police can shut off access to campus buildings (including laboratories)
remotely. One interviewee stated that their institution could immediately shut off building access, at any
time, to anyone. Other institutions stated that they could revoke access to the BSAT laboratories within
hours if the situation necessitated.
Tier 1 BSAT Regulations require a number of additional physical security measures in addition to those
required for Select Agents and Toxins.
A Tier 1 facility must have three security barriers to delay malicious actors. One barrier must be
monitored to detect “intentional and unintentional circumventing of established access control measures
under all conditions (day/night, severe weather, etc.)” and the final barrier must have some form of access
control to ensure that only individuals registered to work with Tier 1 Select Agents are allowed to
pass.2890 Further, procedures must be in place to ensure that powered access control systems maintain
continuity of security in the event of a power disruption.2891 Finally, if the facility is unable to maintain a
security force response time at or under 15 minutes, it must have barriers “sufficient to delay
unauthorized access until the response force arrives,” and therefore be able to prevent “theft, intentional
release, or unauthorized access” to all Tier 1 Select Agents.2892
Entities with Tier 1 Select Agents are further mandated to restrict access to the laboratory and storage
facilities outside of normal business hours by requiring an explicit permission from the facility’s
2889 42 CFR 73.11 (c)(4)-(8), (d)(7)(i)-(v). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2890 42 CFR 73.11 (f)(4)(iv). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2891 42 CFR 73.11(f)(4)(vii). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2892 42 CFR 73.11(f)(4)(viii)(B). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”

responsible official for off-hours access to such facilities.2893 This access regulation formalizes a good
practice introduced in the non-select agent section above.
Physical security at Tier 1 Select Agent facilities is an area of strength in the current regulatory
framework. The regulations provide specific metrics the facility must meet, such as the need for a
minimum of 3 physical barriers with certain requirements, and a maximum 15 minutes security response
time or security barriers adequate to hold off malicious actors until help arrives.2894 Very detailed
guidance has been generated (including a security risk assessment algorithm) to help laboratory managers
assess and mitigate security risks.2895 The regulations are flexible in that laboratories are allowed to
determine, with the help of the relevant security providers, what barriers are appropriate to hold off
potential malicious actors until help arrives. At the same time, the lab’s desired implementation is kept in
check through the required licensing process, whereby CDC or APHIS consider the proposed security
plan before the facility is allowed to work with a Tier 1 Agent.
Research institutions that are registered for Tier 1 BSAT have at least three barriers in place to ensure
regulatory compliance. Some institutions employ more than three barriers. Access to these barriers can be
controlled electronically, physically, by humans, or physiologically.
• Current regulatory and guidance documents do not prohibit use of certain insecure physical
barriers for non-BSAT laboratories. For example, physical barriers that use simple mechanical
keys are insecure and would not necessarily prevent an unauthorized individual from gaining
access to a laboratory.2896 Guidance documents strongly discourage mechanical key locks, and
security planning at the Select Agent and Tier 1 Select Agent levels would presumably prevent
such setups by arguing that these barriers would not noticeably slow an attacker armed with as
little as a crowbar. Discouraging use of inadequate access control measures, such as simple
mechanical locks, for all high containment laboratories could help address the gap.
16.16.4 Surveillance and Monitoring
Surveillance and monitoring measures can be used to detect events such as unauthorized entry, exposure
to infectious agents, and malfunctioning safety and security equipment. Successful surveillance and
monitoring measures enable timely notification of relevant response authorities.
2893 42 CFR 73.11 (f)(4)(ii). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2894 42 CFR 73.11 (f)(4)(iv), (viii)(B). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
CFR Part 331, 9 CFR Part 121, 42 CFR Part 73”;
Jennifer Gaudioso, Susan A. Caskey, LouAnn Burnett, Erik Heegaard, Jeffery Owens, Philippe Stroot, “Strengthening Risk
Governance in Bioscience Laboratories.”

Requirements
Security-specific surveillance and monitoring measures are not required by regulations for all
laboratories.2897 According to the BMBL, “for laboratories not handling select agents, the access controls
and training requirements specified for BSL-2 and BSL-3 in [the] BMBL may provide sufficient
security.”2898
Occupational health monitoring plans are only required for laboratories at the BSL-4 or ABSL-4 levels
(see Figure 16.4).2899 However, biosafety standards make clear that heath surveillance programs are to be
put in place if needed based on the type of work conducted at the facility at any level apart from BSL-1
(including ABSL-1). Health surveillance plans, although typically classed as biosafety measures, also
have an important biosecurity function in detecting a potential exposure incident. Although the health
surveillance program is not designed to discern between deliberate and accidental infections, it would
initiate an isolation process, if necessary, and help mitigate the spread of the disease.
Detection of malfunctioning equipment can prevent the occurrence of an incident, or failing this, at least
minimize its consequences. Thorough equipment checks are typically conducted once a year during
facility shut down. For instance, current OSHA interpretation of regulations require that biosafety
cabinets, which play a crucial role in preventing laboratory infection, must be certified when installed,
when moved, and at least annually.2900
Video surveillance cameras are sometimes present on campuses, at laboratory entrance and exits, in
laboratories not working on Select Agents. Although video surveillance is an oft-cited example of a
surveillance method, most laboratories do not have the staff nor the budget to monitor video feed in real-
time.2901 Laboratories with video surveillance generally use it for logging purposes only and to assist
incident (and accident) investigations. For example, if suspicious activity by an insider is suspected, video
logs of their time in the lab can be retrieved and inspected.
Institutions supporting research in high containment laboratories have developed plans for identifying
potential exposures and contacting the appropriate health authorities.
Shortly after the Virginia Tech shooting in the mid-2000s, most universities established threat assessment
teams to assess potential threats on campus and identify the appropriate approaches and individuals to
address the threat. Several institutions have protocols and reporting mechanisms in place to reports
p. 104-105.
p. 104-105.
p. 120;
42 CFR 73.12(d). U.S. Government Publishing Office, “Title 42: Public Health, §73.12 Biosafety.”
2900 OSHA’s interpretation, based on 29 CFR 1030(e)(2)(iii)(B). U.S. Government Publishing Office, “Title 29: Labor,
§1910.1030 Bloodborne pathogens,” <http://www.ecfr.gov/cgi-
bin/retrieveECFR?gp=&SID=3c86911ce71d159de28dea2738f1d687&r=SECTION&n=se29.6.1910_11030>;
OSHA, “OSHA Fact Sheet- Laboratory Safety Biosafety Cabinets (BSCs),”
<https://www.osha.gov/Publications/laboratory/OSHAfactsheet-laboratory-safety-biosafety-cabinets.pdf>.

incidents that raise concerns. In some cases, these concerns are communicated to the threat assessment
teams directly. If a credible or significant threat presents itself, the threat assessments teams can
communicate directly with the FBI.
Requirements
Surveillance cameras are not explicitly included in the Biological Select Agents and Toxins Regulations.
If installed, they must be described in the security plan, and do not replace visitor escorts.2902
Occupational health monitoring plans are not explicitly included in the Biological Select Agents and
Toxin Regulations for any BSAT, including Tier 1 Select Agents. Occupational health monitoring is
pathogen-specific, and authoritative guidance exists for specific pathogens.2903
Research institutions have installed cameras to monitor access to the BSAT laboratories. This footage is
reviewed periodically by authorized institutional administration and security officials.
Several institutions have armed guard patrols to monitor the perimeter of the facility to identify potential
security concerns. In addition, if an alarm is triggered at a few of the institutions,
Institutions supporting research in high containment laboratories have employee health monitoring
processes and programs regardless of whether the facility works with Select Agents.
Buildings that house animal research and BSAT research conduct perimeter surveillance to identify
possible malicious actors. These surveillance efforts are sometimes real-time and involve police patrolling
the building or involve periodic review of archived surveillance footage.
Requirements
An intrusion detection system must be placed in all places that house or work with Tier 1 Select Agents or
that “reasonably afford access” to such spaces, unless these zones are physically occupied.2904
Tier 1 BSAT Regulations specify that all individuals with access to Tier 1 Select Agents must be enrolled
in an occupational health program.2905
Because none of the laboratories we visited were Tier 1 Select Agent laboratories, no specific information
was collected on surveillance and monitoring of Tier 1 facilities. However, surveillance efforts would be
more stringent than what is currently implemented for BSAT laboratories.
2902 Federal Select Agent Program, “Security Guidance for Select Agent or Toxin Facilities,”
<http://www.selectagents.gov/resources/Security_Guidance_v3-English.pdf>.
2903 42 CFR 73.12(d). U.S. Government Publishing Office, “Title 42: Public Health, §73.12 Biosafety”;
Centers for Disease Control, “Appendix F6 – Guidelines for Medical Surveillance of Laboratory Personnel Working with SARS-
CoV,” <http://www.cdc.gov/SARS/guidance/F-lab/app6.html>.
2904 42 CFR 73.11(f)(4)(v). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2905 42 CFR 73.12(d). U.S. Government Publishing Office, “Title 42: Public Health, §73.12 Biosafety.”

• The length of time between review of footage from video surveillance could prevent rapid
identification, prevention, or response to an incident involving unauthorized access or an actual
event. However, the effectiveness of real-time video may not be significant. 2906
16.16.5 Storage, Inventory, and Accountability Processes
Inventory and material management processes allow labs to keep track of biological materials. Keeping
track of these materials enables loss or theft detection, which can assist in post-event investigations.
Requirements
The BMBL recommends some form of “inventory or material management process for control and
tracking of biological stocks or other sensitive materials” as part of a general biosafety program.2907
Some research institutions routinely check that all pathogen stocks are accounted for. Several high
containment laboratories keep records of stored pathogens. Some institutions lock freezers used to store
their pathogen stocks.
Several laboratories do not do not allow live or active pathogens from being removed from high
containment without adequate fixation, inactivation, or decontamination.
Requirements
Inventory control measures and procedures for reporting and responding to the detected alteration of
inventory records must be codified in the security plan required as part of the Select Agents registration
process.2908 Select Agent regulations stipulate that detailed inventory records be kept for each Select
Agent held in long-term storage, including the name, number of containers, storage location, and chain-
of-custody information.2909 This does not apply to working stocks (i.e. less than 30 days, like inoculated
cells or aliquots diluted to working concentration and intended for use in the near future.2910 Working
2906 Even the DOD 2009 report, which stressed the value of “enhanced and increased video monitoring of the labs,” did not call
for continuous real-time video surveillance. p.20-23, Department of Defense, Defense Science Board Task Force,
“Department of Defense Biological Safety and Security Program,” May 2009, Unclassified, Cleared for Public Release,
http://www.acq.osd.mil/dsb/reports/ADA499977.pdf; National Research Council of the National Academies, Prudent
Practices in the Laboratory: Handling and Management of Chemical Hazards, Updated version, p. 258.
p. 105.
2908 42 CFR 73.11(c)(1), (c)(6), (d)(7)(v). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2909 Volumes must be recorded for toxins only, not agents;
42 CFR 73.17(a)(1), (a)(5). U.S. Government Publishing Office, “Title 42: Public Health, §73.17 Records,”
<http://www.ecfr.gov/cgi-bin/retrieveECFR?r=PART&n=42y1.0.1.6.61>.
2910 CDC, APHIS, “Guidance on the Inventory of Select Agents and Toxins,” p. 6,
<http://www.selectagents.gov/resources/Long_Term_Storage_version_5.pdf>.

stocks must be kept in an access controlled area registered with the Federal Select Agent Program.2911 An
inventory of infected or exposed animals also must be recorded for use in case of escape or theft; the
recorded information includes the quantity, species, location, and final disposition of the animal.2912
The Select Agent regulations require routine inventory audits of pathogens in long-term storage.
Inventory audits must be performed when a collection of Select Agents is moved, when a principal
investigator working with Select Agents leaves or joins the lab, and in the event of theft or loss (at which
point all stocks of agents under the principal investigator responsible for the missing stock are to be
audited).2913
In addition to pathogen accounting, entry and exit to areas holding Select Agents must be recorded,
including the name of the individual, the name of their escort if applicable, and the date and time of
entry.2914 Furthermore, all records stipulated by the Select Agents regulations must themselves have
“controlled access” and must be in such a form that “their authenticity may be verified.”2915
Under current Select Agent regulation 42 CFR 73.11(d)(3), a lock box system is explicitly suggested,
alongside card systems, as a means of meeting the requirement for “freezers, refrigerators, cabinets, and
other containers where select agents or toxins are stored to be secured against unauthorized access.”2916
The BSAT Regulations explicitly allow lock and key systems as a means of ensuring access control to
long-term pathogen storage.2917 Unlike lock boxes, many electronic systems (numeric, card, and
biometric) automatically log the user and the time of access. Electronic logging of such information
would help detect anomalous behavior, such as the opening of a container by an individual at abnormal
hours, and assist in investigating incidents by having an (additional) electronic record of everyone who
accessed Select Agents. These systems enhance an institute’s capability to keep required records
regarding select agent stocks in long-term storage and “information about all entries into areas containing
select agents or toxins.”2918
Institutions that support BSAT research adhere to the federal guidance on long-term storage of BSAT.
Some of the institutions use an automated inventory system where all vials have a bar code. Others secure
pathogens in boxes with security tape to know which boxes have been touched. The freezers are locked.
In addition, several institutions keep paper inventory logs.
Institutions varied in their inventory checks. Some conducted checks on a routine cycle, while others
conducted random inventory checks in addition to the periodic checks. Institutions assess inventory if a
vial appears to be missing.
2913 42 CFR 73.11(e)(1)-(3). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2914 42 CFR 73.17(a)(5). U.S. Government Publishing Office, “Title 42: Public Health, §73.17 Records.”
2915 42 CFR 73.17(a)(7)(b). U.S. Government Publishing Office, “Title 42: Public Health, §73.17 Records.”
2916 42 CFR 73.11(d)(3). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security,”
<http://www.ecfr.gov/cgi-bin/text-
idx?SID=94bd3a730b8387eb15bc058bc4637627&mc=true&node=se42.1.73_111&rgn=div8>.
2917 42 CFR 73.11(d)(3). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2918 42 CFR 73.17(a)(1), (a)(5). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Records,”
<http://www.ecfr.gov/cgi-bin/retrieveECFR?r=PART&n=42y1.0.1.6.61>.

Requirements
No additional storage, inventory, and accountability processes are required for Tier 1 Select Agents that
go beyond those stipulated BSAT.2919 However, Tier 1 Select Agents regulations contain a general clause
that: “an entity's Responsible Official will coordinate their efforts with the entity's safety and security
professionals to ensure security of Tier 1 select agents and toxins and share, as appropriate, relevant
information.”2920
was collected on surveillance and monitoring of Tier 1 facilities.
• Inventory measures and audits facilitate detection of discrepancies and use patterns that may
indicate theft. However, no practical ways exist to measure and track working stocks. No
practical methods exist that would provide accurate working stock pathogen inventory data.
Required accountability checks verify container counts, but recording volume or pathogen
concentrations is not required. Volumes and pathogen concentrations are often recorded by
researchers for experimental purposes, but this is not part of the traceable inventory process.
Practitioners have repeatedly argued that, “beyond knowing who has what pathogen, exact
inventory rules are not informative or feasible, particularly for pathogens actively being
experimented [upon].”2921,2922 The inability to maintain an accurate inventory of working stocks
cannot be resolved. Therefore, working stock control must rely on physical security and
personnel reliability.
16.16.6 Transfer, Shipment, and Chain-of-Custody Protocols
Secure transfer of pathogens involves: 1) ensuring proper documentation and approvals are provided; 2)
not alerting anyone to the contents of the package during shipment, if relevant; and 3) comprised of
means to detect, report, and respond to missing or damaged packages. Other than clinical and diagnostic
samples, pathogens are rarely shipped. In addition, recent incidents of accidental shipment of live or
incorrect pathogens have resulted in only a few transportation companies willing to ship infectious agents.
2919 U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2920 42 CFR 73.11 (f)(2). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
2921 AAAS, AAU, APLU, FBI, Bridging Science and Security for Biological Research: Implementing the Revised Select Agents
and Toxin Regulations Proceedings from the Meeting, p. 15.
2922 Nancy Connell, “Biological Agents in the Laboratory- The Regulatory Issues

Requirements
Department of Transportation regulations categorizes infectious substances as “Division 6.2” goods for
shipment under transport regulations, and further divided into two categories (A and B).2923 Category A is
for an “infectious substance in a form capable of causing permanent disability or life-threatening or fatal
disease in otherwise healthy humans or animals when exposure to it occurs,” whereas Category B is for
any other infectious substance.2924 Classification of an infectious substance as Category A or B “must be
based on the known medical history or symptoms of the source patient or animal, endemic local
conditions, or professional judgment concerning the individual circumstances of the source human or
animal.”2925
Both Category A and B substances are triple packaged for transport, although Category A packaging
requirements are more stringent in terms of leak resistance.2926,2927 Air shipment requires that at least one
side of the external layer be emblazoned with an “Infectious Substances” diamond hazard
label.2928,2929,2930,2931 Although transportation regulations require a security transport plan for a hazardous
goods shipment, “Division 6.2” goods are not covered by these regulations.2932
CDC and USDA require permits to import pathogens.2933,2934,2935 USDA also requires a permit to ship
2923 U.S. Government Publishing Office, “Title 49 Transportation, §173.134 Class 6, Division 6.2 – Definitions and exceptions,”
idx?SID=c43d9605516b239af6c12f288eef1a86&mc=true&node=se49.2.173_1134&rgn=div8>. Similar categorization of
agents is included in the International Air Transport Association’s Dangerous Good Regulations.
2924 49 CFR 173.134 (a)(1)(i)-(ii). U.S. Government Publishing Office, “Title 49 Transportation, §173.134 Class 6, Division 6.2
– Definitions and exceptions.”
2925 U.S. Government Publishing Office, “Title 49 Transportation, §173.134 Class 6, Division 6.2 – Definitions and exceptions.”
2926
For instance, if shipped at ambient temperatures or higher, Category A substances must have a positive means of ensuring a
leakproof seal.
U.S. Government Publishing Office, “Title 49 Transportation, 173.196 Category A infectious substances,”
idx?SID=c43d9605516b239af6c12f288eef1a86&mc=true&node=se49.2.173_1134&rgn=div8.
2927 U.S. Government Publishing Office, “Title 49 Transportation, 173.199 Category B infectious substances,”
idx?SID=8dd11b1ac9a22c45afc96ac5b2489afe&mc=true&node=pt49.2.173&rgn=div5#se49.2.173_1199>.
2928 U.S. Department of Transportation, Pipeline and Hazardous Materials Safety Administration (PHMSA), “Transporting
Infectious Substances Safely,” October 1, 2006, listed as of August 2015 as current on PHMSA website,
<http://www.phmsa.dot.gov/pv_obj_cache/pv_obj_id_54AC1BCBF0DFBE298024C4C700569893C2582700/filename/Tran
sporting_Infectious_Substances_brochure.pdf>,
2929 http://phmsa.dot.gov/portal/site/PHMSA/menuitem.6f23687cf7b00b0f22e4c6962d9c8789/?-
vgnextoid=4d1800e36b978410VgnVCM100000d2c97898RCRD&vgnextchannel=0f0b143389d8c010VgnVCM100000804
9a8c0RCRD&vgnextfmt=print.
2930 UPS, “Infectious Substances, Category A,”
<http://www.ups.com/content/us/en/resources/ship/hazardous/responsible/diagnostic.html>;
2931 University of Virginia, “Shipping Infectious Substances by Air,” <http://ehs.virginia.edu/biosafety/bio.transport.air.html>
2932 U.S. Government Publishing Office, “Title 49 Transportation, 172 Subpart I- Safety and Security Plans,”
idx?SID=1530a4d53604eb266607b121832fd2d2&mc=true&node=sp49.2.172.i&rgn=div6>
2933 CDC issues permits for human pathogens and USDA issues permits for animal and plant pathogens.
2934 9 CFR 122, 42 CFR 71.54. U.S. Government Publishing Office, “Title 9 Animals and Animal Products, Part 122 Organisms
and Vectors,” <http://www.ecfr.gov/cgi-bin/text-
idx?SID=fb98f5f1a14891a6cbd139179bc3dfae&mc=true&node=pt9.1.122&rgn=div5>;
2935 U.S. Government Publishing Office, “Title 42 Public Health, Part 71 Foreign Quarantine, Subpart 71.54 Import regulations
for infectious biological agents, infectious substances, and vectors,” <http://www.ecfr.gov/cgi-
bin/retrieveECFR?gp=1&SID=e170ce9bdb27d491a0e1d31d7bffeb2f&ty=HTML&h=L&r=SECTION&n=42y1.0.1.6.59.6.1
9.5%20%28>

pathogens across state lines, and CDC sometimes requires a permit to transfer imported pathogens across
state lines.2936 CDC also provides detailed instructions on safe packaging of infectious substances for
shipment.
Similarly, international export of biological agents may require a Department of Commerce export control
permit if they are restricted and not exempt. In addition, a Department of State registration and permit
may be needed if the agent falls within the ITAR regulations for arms control.2937,2938,2939,2940 Commerce
regulations apply to pathogens restricted by the Australia Group and Select Agents and State regulations
apply to biological agents “modified to increase…capability to produce casualties in humans or
livestock.”2941
Institutions appear to require hazardous materials shipping training and certification for all employees
who are designated as shippers.2942 In addition, institutions have offices dedicated to ensuring compliance
with all export control regulations. Some institutions have a designated individual, “Export Controls
Coordinator,” to provide assistance with the export control requirements.2943
Requirements
The transfer of BSAT between separate entities licensed to possess BSAT requires prior approval by
either CDC or APHIS unless the Select Agent is contained in a specimen for proficiency testing. In the
latter case, CDC or APHIS must simply be informed at least seven calendar days prior to the transfer.2944
For all transfers of BSAT, the recipient must keep records of shipments and report to CDC or APHIS
within 48 hours after the slated delivery time that the shipment has been received as planned, or that it is
delayed or missing.2945 Furthermore, the recipient must immediately report to CDC or APHIS if the
package was damaged to the point that a release may have occurred.2946
2936 U.S. Government Publishing Office, “Title 42 Public Health, Part 71 Foreign Quarantine, Subpart 71.54 Import regulations
for infectious biological agents, infectious substances, and vectors.” Center for Disease Control and Prevention. Interstate
Shipment of Etiologic Agents. Accessed on http://www.cdc.gov/vaccines/pubs/surv-manual/appx/appendix24-etiologic-
agent.pdf. Accessed on September 19, 2015.
2937 Usually “fundamental research” is exempted from the Commerce permits, but not in the cases of biological weapons
potential or restricted publication of results.
2938 15 CFR 734.3-8, U.S. Government Publishing Office, “Scope of the Export Administration Regulations,”
<http://www.gpo.gov/fdsys/pkg/CFR-2001-title15-vol2/pdf/CFR-2001-title15-vol2-part734.pdf>;
2939 15 CFR 744.4-6, U.S. Government Publishing Office, “Control Policy: End-User and End-Use Base,”
<http://www.gpo.gov/fdsys/pkg/CFR-2001-title15-vol2/pdf/CFR-2001-title15-vol2-part744.pdf>
2940 22 CFR 121.1(XIV)(b) U.S. Government Publishing Office, “The United States Munitions List,” <http://www.ecfr.gov/cgi-
bin/text-idx?SID=88e7fab9254a3319c3df6fb11a2233ab&mc=true&node=se22.1.121_11&rgn=div8>
2941 U.S. Department of Commerce, Commerce Control List, “Category 1 – Special Materials and Related Equipment,
Chemicals, ‘Microorganisms’ and ‘Toxins’,” <http://www.bis.doc.gov/index.php/forms-documents/doc_download/989-
ccl1>.
2942 For example: University of California, Irvine, Environmental Health and Safety, “Shipper’s Responsibilities,”
<http://www.ehs.uci.edu/programs/dgoods/>.
2943 For example: University of Colorado Boulder, Office of the Vice Chancellor for Research, Research Administration and
Support, “ORI (Compliance), Export Controls, Guidance, Biological Agents,” <http://www.colorado.edu/vcr/export-
controls/guidance/biological-agents>.
2944 The transfer request is made using APHIS/CDC Form 2. U.S. Government Publishing Office, “Title 42: Public Health,
§73.16 Transfers.”
2945 42 CFR 73.16(j). U.S. Government Publishing Office, “Title 42: Public Health, §73.16 Transfers.”
2946 Ibid.

The same Category A and B packaging rules described above apply for all pathogens. That is, a package
carrying a Select Agent is visually indistinguishable from a package carrying a pathogen not on the Select
Agent list.
Finally, Select Agent regulations have a clause regarding suspicious packages. Any suspicious packages
must be inspected “before they are brought into or removed from the area where select agents or toxins
are used or stored.”2947 The rationale behind this regulation is that a suspicious package may contain an
explosive device, which could then potentially breach containment.2948
All packages are prepared by a BSAT approved individual, transferred to the shipper in person, and
received from the shipper in person by a BSAT approved individual. Chain-of-custody is maintained
throughout. Once a package is received by the recipient institution, its outer packaging is examined for
any damage before it is transported to the recipient laboratory’s containment facility wherein the
package’s contents will be examined for any damage. The CDC and shipper are notified immediately if
the package is damaged.
Transportation security measures must balance the desire for additional security measures against the
desire to avoid drawing attention to a particular shipment. Select Agent shipments that are visually
indistinguishable from any other shipments of infectious substances once packaged limits the risk of
highlighting the package with the restricted BSAT. Knowing the dates of shipment, the identification of
the exact trucks carrying the virus, and transportation lines used is impossible without access to the
shipping and tracking information.
Very few pathogen shipments occur each month, which is confirmed by the CDC and APHIS joint
informational website, which provides additional information on transfer frequency and security. The
webpage on BSAT states that “approximately 4250 transfers that have occurred since 2003,” with “one
confirmed loss of a select agent that occurred during shipment.”2949 The FBI investigation that resulted
“determined that the loss most likely did not occur at either the shipping or receiving areas,” i.e. that the
package was apparently lost during the transit portion itself.2950 Furthermore, GoF viruses apparently are
not shipped.
Interviewees further noted that the Department of Transportation did surprise inspections to ensure that
transfers of pathogens were conducted according to the regulations.
2947 42 CFR 73.11(d)(4). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”
CFR Part 331, 9 CFR Part 121, 42 CFR Part 73,” July 5, 2013, p.23-24, 38,
<http://www.selectagents.gov/resources/Security_Guidance_v3-English.pdf>.
2949 CDC, APHIS, “Federal Select Agent Program Guidance on the Shipment and Receipt of Packages with Select Agents and
Toxins,” http://www.selectagents.gov/guidance-shipreceipt.html.
Toxins.”

Requirements
No additional regulations exist for transport of Tier 1 Select Agents beyond those applicable for all
BSAT.2951 The same Category A and B packaging rules described above apply for all pathogens. As a
result, a package carrying a Tier 1 Select Agent should be visually indistinguishable from one carrying a
Select Agent pathogen not also a Tier 1 Select Agent, or one carrying a pathogen that is not a Select
Agent.
was collected on surveillance and monitoring of Tier 1 facilities.
16.16.6.4 Gap Analysis
• Best practices for transportation security are not public, if they exist. General methods to mitigate
transportation vulnerability include ensuring that the transport has GPS tracking and a transport-
based alert system that contacts police in case of an emergency (readily available in retail and
armored transport vehicles).2952,2953 Another vulnerability-reducing approach is to ensure that the
transporting company monitors have the appropriate points of contact to quickly relay
information to the appropriate law enforcement agency. Providing an approach through which
practitioners can share best practices could enhance transportation security.
16.16.7 Emergency Response Protocols
Emergency response plans, drills, and notification systems prepare research facilities to respond to all-
hazards emergencies, including security emergencies.
Requirements
All laboratories should have a general emergency plan as part of their OSHA worker safety requirements,
at the very least to deal with fires and with natural emergencies (such as earthquakes, tornadoes, or
floods).
Toxins.”
2952 For examples in common use, see: Rory Reid, “Citroen eTouch emergency panic button calls cops automatically,” CNET,
October 5, 2010, <http://www.cnet.com/news/citroen-etouch-emergency-panic-button-calls-cops-automatically/>; “Avis
`Panic Button’ Debuts in Miami Cars,” Orlando Sentinel, September 14, 1994, <http://articles.orlandosentinel.com/1994-09-
14/business/9409130599_1_guidestar-avis-emergency-button>.
2953 For features commercially available in the high-security transport field, see for example: 3SI Security Systems, “Cash-in-
Transit (CIT) Tracker™,” <https://www.3sisecurity.com/products/en-cash-in-transit-cit-tracker/>

Concerns about active shooters on the institution’s property and the consequences of natural disaster,
research institutions plan and conduct large and small-scale exercises to practice response and identify
potential areas for improvement. Small-scale exercises include relevant members of the institution
whereas large-scale exercises involve local police and first responders and FBI. Institutions practice a
wide range of exercises to make sure that the appropriate institutional officials know what to do and with
whom to communicate in an emergency situation.
In addition, all building, electrical, and safety equipment are tested periodically.
Some institutions had emergency operations centers to facilitate communication and coordinate response
efforts in an emergency.
16.16.7.2 Additional measures at the select agent level
Requirements
Annual drills are required to test emergency and incident response plans.2954
An effective emergency response depends on appropriate planning to ensure that the response is
coordinated and appropriate for the situation, lines of communication with the laboratory to ensure the
safety of laboratory personnel, appropriate equipment, and familiarity with using said equipment and the
laboratory layout. Given the complexity of responding to security situations at a high-containment
laboratory, law enforcement is actively involved with laboratory-organized security training exercises.
All BSAT institutions involve laboratory staff to practice responses to small-scale incidents that occur in
the laboratory, such as spills. These small-scale exercises are sometimes conducted a few times a year. In
addition, several institutions conduct medium-sized exercises featuring rotating scenarios with
institutional or local law enforcement and first responders to ensure all individuals are property trained to
respond to different types of emergencies.
Requirements
Entities with Tier 1 Select Agents must have a security response time at or below 15 minutes, or
otherwise provide barriers that are “sufficient to delay unauthorized access until the response force
arrives.”2955 A facility’s security response time is measured starting from the tripping of an intrusion
alarm or incident report, to the arrival of the security force to the first security barrier.
The institutions that support Tier 1 BSAT conduct exercises with institutional and/or local law
enforcement and first responders to test response activities and ensure that all individuals have the proper
information and training to safely respond to emergency situations.
2954 42 CFR 73.14(f). U.S. Government Publishing Office, “Title 42: Public Health, §73.14 Incident response.”
2955 42 CFR 73.11(f)(4)(viii)(A). U.S. Government Publishing Office, “Title 42: Public Health, §73.11 Security.”

We did not identify any gaps in emergency response. Based on our discussions, institutional, local, and
federal law enforcement and relevant institutional officials appear to identify productive ways of working
together in different scenarios.
However, we identified a gap when discussing exercises and drills to practice emergency response. The
degree to which facilities conduct exercises on security-related incidents varies. Exercise topics include
response to fires and bombs, natural disasters, and biosafety incidents, such as spills. The nature of the
exercise planning process and the participation of local first responder agencies varies by exercise and
facility.
16.16.8 Indirect Security Measure: Screening Framework Guidance for Providers of Synthetic
Double-Stranded DNA
Although the Screening Framework Guidance for Providers of Synthetic Double-Stranded DNA does not
require security practices to be implemented at research institutions, it is included in this governance
section as an indirect measure. Concerns about development of biological agents using chemical synthesis
of viral genomes have driven the development of this Guidance. GoF viruses are generated in the
laboratory using genetic engineering and, in some laboratories, synthetic genomics.
Concerns about chemical synthesis of pathogen genomes and the ability to purchase virulence genes and
genes from Biological Select Agents and Toxins raised significant concern among the security policy
community in the early 2000s. During this time, the U.S. government and international community were
evaluating the potential for life science and biotechnology to enable both beneficial and destructive
research and two research groups published scientific articles on chemical synthesis of infectious human
and bacterial viruses, both of which were carried out using DNA molecules purchased from DNA
synthesis providers.2956 The publication of these papers led to concerns within the security and security
policy communities about creation of viral genomes, particularly of viruses on the Biological Select
Agents and Toxins list. These concerns prompted the NSABB to evaluate the biosecurity considerations
associated with synthesis of Biological Select Agent Toxins and recommend approaches to address any
risks.2957 Three of NSABB’s recommendations were:
1. The Departments of Health and Human Services and Agriculture develop “harmonized guidance
to investigators and nucleic acid/gene/genome providers concerning the SAR [Select Agent
Regulations] with respect to synthetically-derived DNA.”
2. The federal government develop a process that providers can use to screen for Biological Select
Agents and Toxins, develop and promote “preferred practices for screening orders and
interpreting results,” among other related activities, and
3. Evaluate current biosafety guidelines to ensure that guidelines and regulations are adequate for
synthetically derived DNA.
2956 At that time, a fairly new industry of gene synthesis providers had developed to provide the service of making genes from
DNA sequences submitted by its customers. The field was enabled by technologies that allow for long pieces of DNA to be
made chemically and with high fidelity.
2957 National Science Advisory Board for Biosecurity. Addressing Biosecurity Concerns Related to the Synthesis of Select
Agents. Dec 2006. Accessible at
http://osp.od.nih.gov/sites/default/files/resources/Final_NSABB_Report_on_Synthetic_Genomics.pdf. Accessed on
September 18, 2015.

In 2010, the Department of Health and Human Services issued its Screening Framework Guidance for
Providers of Synthetic Double-Stranded DNA.2958 This is a voluntary guidance that includes provisions
for screening of sequences and customers. The overarching goal of the Guidance is to “minimize the risk
that unauthorized individuals or individuals with malicious intent will obtain “toxins and agents of
concern” through the use of nucleic acid synthesis technologies, and to simultaneously minimize any
negative impacts on the conduct of research and business operations.” The Guidance states that
companies should screen customers to verify their identity and legitimacy, identify any “red flags,” and
ensure all U.S. trade and export control regulations are followed. Sequence screening involves evaluating
the order sequence to determine whether it is more similar to a sequence from a Biological Select Agent
or Toxin than it is to a sequence from an organism not on that list. If it is, the Guidance states that the
company should conduct follow-up screening to “verify” that the customer has a legitimate use of the
gene and “is acting within their authority.” In addition, the Guidance provides resources to gene synthesis
providers to assist in consulting the appropriate U.S. regulations or guidance, contact the Federal Bureau
of Investigation Weapons of Mass Destruction Unit if any concerns arise, and consult with the Select
Agent Program and Department of Commerce if questions arise.
In practice, the gene synthesis industry has changed since the Guidance has been released. A series of
commercial acquisitions have changed the landscape of gene synthesis companies where many of the
companies engaged in the development of the NSABB recommendations and U.S. government guidance
have been consolidated. Other companies, not previously engaged seem to have appeared in this space. In
addition, companies from China seem to have become engaged in the international consortiums for gene
synthesis companies. At least two international industry associations have emerged and both have
discussed and encouraged their members to screen sequences and customers. Members of the
International Gene Synthesis Consortium (IGSC) have formed a non-profit corporation to make it easier
for small companies, non-profit organizations, and academic institutions to “leverage the biosecurity
expertise of the IGSC.”2959
16.16.9 Governance of Hazardous Chemicals
Life science research often involves use of hazardous chemicals. Many of these chemicals are toxic and
some are flammable, reactive, or explosive.2960 These chemicals could be misused by a malicious actor to
facilitate other malicious acts, including arson, bombing, and sabotage.
Regulations and best practices governing the storage and use of hazardous chemicals limits the ability of
malicious actors to divert hazardous supplies already present within the laboratory to carry out malicious
acts. Hazard communication regulations require the labelling of hazardous chemicals, and stipulate that
employees must be made aware of chemical hazards through training.2961
Regulations explicitly require that laboratories must “minimize all chemical exposures and risks.”2962
Chemical risk mitigation is done at the facility level through a Chemical Hygiene Plan, which specifies
2958 Department of Health and Human Services. Screening Framework Guidance for Providers of Synthetic Double-Stranded
DNA. 2010. Accessible at http://www.phe.gov/Preparedness/legal/guidance/syndna/Documents/syndna-guidance.pdf.
2959 Schubert E. International Gene Synthesis Consortium Forms Not-for-Profit Corporation. April 28, 2015. PRLOG.
Accessible at http://www.prlog.org/12450359-international-gene-synthesis-consortium-forms-not-for-profit-
corporation.html. Accessed on September 18, 2015.
Chemical Hazards, Updated version. p. 53-74.
2961 U.S. Government Publishing Office, “Title 29 Labor, §1910.1200 Hazard Communication,” <http://www.ecfr.gov/cgi-
bin/text-idx?rgn=div8&node=29:6.1.1.1.1.1.1.36>.
2962 U.S. Government Publishing Office, “Title 29 Labor, §1910.1450 Occupational exposure to hazardous chemicals in
laboratories,” <http://www.ecfr.gov/cgi-bin/text-idx?rgn=div8&node=29:6.1.1.1.1.1.1.36>.

what measures the employer will take to protect the employees from chemical hazards.2963 Regulations
further require that any work with hazardous chemicals must be preceded by a risk assessment, which will
“identify chemicals to be used, amounts required, and circumstances of use in the experiment.”2964 The
National Research Council of the National Academies (NRC), funded by NIH, has provided extensive
guidance on minimizing chemical hazards.2965
In addition, institutions possessing sufficient quantities and types of chemicals that are covered by the
Chemical Facility Anti-Terrorism Standards must also comply with its personnel, physical, and inventory
security requirements. However, many universities are exempt from these requirements because they do
not possess the minimum quantity of chemicals as stipulated in the Standards.
16.16.10 Gaps in Security Governance
In addition to gaps described in the previous section, the following overarching issues were identified:
16.16.10.1 Financial and Technical Resources
The level of financial and technical resources made available to maintain Select Agent facilities at a high
security level in light of stricter regulations is of significant concern. Regulations have become stricter to
meet growing security concerns. At the same time, few additional financial, administrative, and
informational resources have been made available for laboratories to meet these new requirements.2966
Institutional administrators have repeatedly raised these issues in light of decreased state funding and
attrition of select agent facility staff.2967 Without sufficient program funds, institution managers will have
to implement cuts elsewhere to meet the minimum regulatory requirements. For instance, the number of
full-time biosafety employees may be reduced.2968 Furthermore, this situation is significantly exacerbated
by the current level of regulatory burden facing research institutions. The only known estimate of cost
burden was produced by the Federal Select Agent Program before the most recent regulatory changes to
the BSAT Regulations.2969 However, to the best of our knowledge, no other assessments that quantify
time spent, financial cost of implementing the regulations, and opportunity costs exist.
16.16.10.2 Lack of Clarity about Requirements
A lack of clarity about effective practices to implement the security requirements exists. In some cases,
such as for personnel security, the Federal Select Agent Program has issued a guidance document.
However, the continuously changing regulatory environment, variability across inspections, and the lack
laboratories.”
laboratories.”
Chemical Hazards, Updated version.
2966 AAAS, AAU, APLU, FBI, Bridging Science and Security for Biological Research: Implementing the Revised Select Agents
and Toxin Regulations Proceedings from the Meeting, p. 13-15.
2967 Ibid.
2968 A practitioner survey conducted in 2008 found that at the BSL-3/ABS-3 laboratory level, more than half (64%) of the
respondents indicated their facility operated with less than three full-time equivalent employees devoted to biosafety.
Allison T. Chamberlain et al., “Biosafety Training and Incident-reporting Practices in the United States: A 2008 Survey of
Biosafety Professionals,” Applied Biosafety 14, no. 3 (2009): p. 138,
<http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2947438/>.
2969 Centers for Disease Control and Prevention and U.S. Department of Agriculture. Regulatory Impact Analysis & Final
Regulatory Flexibility Analysis. 2011. Accessible at http://system.suny.edu/media/suny/content-
assets/documents/compliance/ehs/Regulatory-Impact-Analysis-and-Final-Reg-Flexibility-Analysis.pdf. Accessed on
September 19, 2015.

of a forum through which to discuss best practices for biological laboratory security, presents a gap
between the requirement, regulators, and regulated community. Efforts to identify and resolve specific
problems to enhance security beyond compliance checklists could help address gaps specific to certain
security measures.
Confusion on regulations is detectable from the lack of consistency between inspection results from the
same facility, where different inspectors interpret existing regulations differently.2970 In one case, a
laboratory was cited for not having provided workers with animal subjects training when the laboratory
did not conduct work with animals.2971 Such uncertainty could lead to laboratory managers dedicating
resources to meeting interpretations of the legislation in an effort to avoid regulatory trouble, resulting in
a negative impact on security-relevant spending. Unclear regulations could also risk seeing objectively
unsatisfactory, but technically correct implementation.
Results from inspections carried out by the Office of the Inspector General of the Department of Health
and Human Services in the 2003-2005 period demonstrated that a significant time gap could exist
between the roll-out of new regulations and satisfactory implementation across all concerned institutes.
These inspections demonstrated gaps in implementation of regulations at some institutes, ranging from
weaknesses in access controls, to insufficient security plans and incident response plans.2972,2973,2974,2975
The current environment of diminished resources and a lack of consensus in regulatory interpretations
would probably slow implementation of any further regulations and potentially impedes current
implementation of regulations.
16.16.10.3 Guidance on Integrating Cross-over Biosafety and Biosecurity Measures
Appropriate guidance was not identified for integrating biosafety and biosecurity measures, such as waste
management systems. Harmonization of guidance for these measures would enhance biological safety
measures to prevent its exploitation and resolve discrepancies between practice and biosafety/biosecurity
objectives. Examples exist where practices sufficient for promoting biosafety may present opportunity for
a malicious actor.
16.16.10.4 Regulatory Nomenclature of Pathogens and Toxins
Nomenclature issues with respect to infectious diseases may lead to confusion. The use by different
branches of the federal government of several tier lists and risk categories for pathogen characterization is
a potential area of confusion, with potential negative repercussions on laboratory compliance. For
example, USDA/APHIS has a list entitled “High-Consequence Foreign Animal Diseases and Pests” in
addition to the more established animal and plant “Select Agents” list. Within both of these lists exists a
subset of pathogens (“Tier 1”) described to present significant security threats. On top of these
2970 Practitioners have argued that inspections “rely heavily on individual inspector interpretations of the regulations.” AAAS,
AAU, APLU, FBI, Bridging Science and Security for Biological Research: Implementing the Revised Select Agents and
Toxin Regulations Proceedings from the Meeting, p. 14-15.
2971 Ibid.
2972 Daniel R. Levinson, Inspector General, Office of the Inspector General, Department of Health and Human Services,
“Summary Report on State, Local, Private, and Commercial Laboratories’ Compliance With Select Agent Regulations (A-
04-06-01033),” January 9, 2008, p. i-ii, <http://oig.hhs.gov/oas/reports/region4/40601033.pdf>;
2973 Daniel R. Levinson, “Summary Report on Universities’ Compliance with Select Agent Regulations (A-04-05-02006),” June
30, 2006, <http://oig.hhs.gov/oas/reports/region4/40502006.pdf>
2974 Dara Corrigan, Acting Principal Deputy Inspector General, “Summary Report on Select Agent Security at Universities (A-
04-04-02000),” May 25, 2004, <http://oig.hhs.gov/oas/reports/region4/40402000.pdf>;
2975 See also the list of cases in the Sandia National Laboratory report SAND2009-8070: Jennifer Gaudioso, Susan A. Caskey,
LouAnn Burnett, Erik Heegaard, Jeffery Owens, Philippe Stroot, “Strengthening Risk Governance in Bioscience
Laboratories,” p.81-94.

designations, separate “Category A and B” designations are used by both the Department of
Transportation and NIH/NIAID are of different composition and used for different purposes despite
similar terminology. Keeping the agents, lists, and designations straight may present challenges to
individuals and institutions who are complying with requirements from several or all of these departments
and agencies.
16.16.11 Major Challenges
The major challenges emerging from the evaluation are:
• Research involving infectious disease and animal research are governed by numerous Executive
Orders, laws, guidance, and contractual requirements at the federal level. In general, this tapestry
of governance appears to be effective at preventing/mitigating physical security risks. However,
of all required security measures, personnel security (i.e., identifying, assessing, and preventing
the insider threat) presents the largest implementation challenge, in part because of the required
processes for vetting employees for Biological Select Agents and Toxins.
o The variability in implementation of security requirements across all research institutions
presents a challenge when considering the effectiveness of federal governance. This
variability results from differences in financial and human resources, lack of standards for
security measures, institutional structure and support, and state, local, and institutional
policies. The institutions that were visited as part of this effort differ significantly from those
institutions that have been highlighted in the popular press as having poor security,
o No set of best practices or validated practices exist for implementation of security

requirements. Best or validated security measures could address concerns about variability in
implementation of security requirements and about inconsistent inspections,
o Security awareness appears to be high among administrators and employees who work with
Biological Select Agents and Toxins and/or research involving animals. However, this
awareness is not necessarily pervasive across the entire life science research community,
o Some security measures, such as personnel security, are governed by other regulations, such
as for restricting access to radioisotopes and certain hazardous chemicals. Some life science
researchers are required to undergo personnel security assessments for work with
radiological materials, chemicals, and Biological Select Agents and Toxins. However, each
requirement and vetting process differs across the regulations and some processes are
viewed are more effective than others,
o One institution suggested the creation of an institutional mechanism through which

the Responsible Official and campus police could share information about potential
concerns, complaints, or arrests of Federal Select Agent Program-approved
individuals,
o In light of increasing cyber breaches in many sectors, innovative technical and policy
options are lacking for securing computer systems that control facility operations and store
or house data (e.g., surveillance footage, digital inventories, and personnel information) from
hacking,
o One of the more significant challenges is keeping pace with the changing social
environment of U.S. research laboratories. Though not evaluated sufficiently in this
effort, the increasingly multidisciplinary, multi-sectoral, international, and digital
research enterprise likely will outpace conventional physical, electronic, and
personnel security measures. However, development, validation, and adoption of
security measures that both counters emerging threats and enable continued growth of
this enterprise has yet to be addressed, and
o The reality that the statutory landscape governing Biological Select Agents and Toxins
undergoes constant change presents difficulties to implementation of effective security
measures, not simply measures to meet compliance requirements. Biosecurity regulations
have become a moving target causing institutions difficulty in implementing the new
requirements before the regulations change again.
• Biosafety measures for restricting personnel access to high containment laboratories, imposing
physical and electronic barriers to restrict unauthorized access and preventing accidental release
of the pathogen, and surveillance and monitoring have a dual purpose of enhancing safety and
contributing to security,
• Of the institutions project staff visited, all implemented measures that either met or exceeded the
federal requirements for security based on evaluating interview responses, measures observed on
site, and compliance with federal requirements,
• Research administrators, and some senior scientists, have open and cooperative relationships with
their institutional police and local FBI WMD Coordinator,
• The intense focus of security on a Biological Select Agents and Toxins results in missed
opportunities for raising awareness of security risks across the entire life science research
enterprise, and
• Significant issues remain, including the availability of adequate resources, consistency and clarity
of security requirements and inspections, and classification nomenclature of pathogen categories.
16.16.12 Knowledge Gaps
In evaluating the security measures required for and implemented at research institutions conducting
research with influenza, SARS-CoV, and MERS-CoV, several knowledge gaps emerged. Addressing
these gaps may enable more comprehensive assessment of the security risks associated with the conduct
of different types of pathogens. However, some of these gaps present a security risk if communicated in
publicly available literature.
Knowledge gaps include:
• Security measures implemented at non-select agent, non-animal research facilities,
• The financial, human, educational, scientific, and security costs involved in implementing
security requirements,
• The existence of standards for training inspectors who assess compliance with security
requirements,

• Best practices for implementing security measures,
• The prevalence of additional physical security measures across all U.S. BSL-3 laboratories not
working with Select Agents is a knowledge gap. In part this is because the NIH Physical Security
Design Requirements only applies to new or refurbished facilities, and the available information
is insufficient to: a) judge how many BSL-3 laboratories are old and have not been refurbished,
and, hence, are not required to meet the NIH physical security design requirements; and b)
determine the difference in physical security between any such old laboratories, and laboratories
meeting the non-public NIH physical security design requirements, and
• Insufficient knowledge regarding the cumulative access delay for physical access barriers for
non-BSAT and BSAT laboratories.
In October 2015, the U.S. Government released recommendations by the Federal Experts Security
Advisory Panel (FESAP) and the Fast Track Action Committee on Select Agent Regulations (FTAC-
SAR) to strengthen biosafety and biosecurity practices and oversight of facilities that conduct BSAT
research.2976,2977,2978 These recommendations span from promoting an environment of security awareness
to establishing a mechanism through which best practices can be shared. Some of these recommendations
address long-time challenges of the regulated community including some highlighted in this report, while
others incorporate approaches taken by U.S. Government agencies as part of their outreach activities.
Following a 90-day internal review of the Centers for Disease Control and Prevention (CDC)/ Division of
Select Agents and Toxins, the CDC issued a report detailing specific recommendations for addressing the
reviewers’ observations on inspections, incident reporting, and transparency and public understanding.2979
The CDC’s observations are consistent with the challenges described in this report and previously
highlighted by regulated research institutions.
2976 U.S. Government. Fact Sheet: Enhancing Biosafety and Biosecurity. October 2015.
2977 U.S. Government. Report of the Federal Experts Security Advisory Panel. December 2014.
2978 National Science and Technology Council, Committee on Homeland and National Security. Fast Track Action Committee
Report: Recommendations on the Select Agent Regulations Based on Broad Stakeholder Engagement. October 2015.
2979 Centers for Disease Control and Prevention. CDC 90 Day Internal Review of the Division of Select Agents and Toxins.
Accessible at: http://www.cdc.gov/phpr/dsat/full-report.htm. Accessed on November 5, 2015.

Risk and Benefit Analysis of Gain of Function Research - Draft Final Report PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Risk and Benefit Analysis of Gain of Function Research - Draft Final Report PDF

Uploaded by

Copyright:

Available Formats

Preface: A Brief Guide to Gryphon’s Risk and Benefit Analysis of Gain of

• Chapter 2. A detailed introduction and description of the purpose of the RBA

Draft Final Report—December 2015

2 Overview and Purpose 8

4 Background Information on Influenza Viruses, the Coronaviruses and GoF Research 16

5 Historical Context of Outbreaks of Influenza, SARS and MERS 42

6 Risk Assessment of Laboratory Accidents and Natural Disasters 76

7 Biosecurity Risk of Malicious Acts Targeting a GoF Laboratory 168

8 Biosecurity Risk of GoF Information 211

9 Benefit Assessment of GoF Research 240

10 Potential Proliferation of GoF Research 445

11 Risk of Loss of Trust in Science 464

12 Appendix I: Glossary 480

13 Appendix II. Acronyms Used 486

15 Appendix IV. Benefit Assessment 518

16 Appendix V: Findings Informing the Biosecurity Risk Assessment 826

1.1 Biosafety Risk Assessment

GoF RBA Report Gryphon Scientific, LLC 1

GoF RBA Report Gryphon Scientific, LLC 2

GoF RBA Report Gryphon Scientific, LLC 3

If well-funded, GoF research could be conducted by up to approximately 40 research groups in the US

1.2 Biosecurity Risk Assessment of Malicious Acts Targeting a GoF Laboratory

GoF RBA Report Gryphon Scientific, LLC 4

1.3 Biosecurity Risk Assessment of GoF Information

GoF RBA Report Gryphon Scientific, LLC 5

1.4 Benefit Assessment of GoF Research

GoF RBA Report Gryphon Scientific, LLC 6

GoF RBA Report Gryphon Scientific, LLC 7

• That an outbreak caused by a laboratory accident may occur,

2.1 Organization of the Project

GoF RBA Report Gryphon Scientific, LLC 8

2.3 Interpreting the Results of the RBA

GoF RBA Report Gryphon Scientific, LLC 9

GoF RBA Report Gryphon Scientific, LLC 10

3.1 RA of accidents and natural disasters

GoF RBA Report Gryphon Scientific, LLC 11

GoF RBA Report Gryphon Scientific, LLC 12

3.2 Biosecurity RA of malicious acts targeting a laboratory

GoF RBA Report Gryphon Scientific, LLC 13

3.3 Biosecurity RA of GoF Information

3.4 Benefit Assessment

1. A foundation for the analysis was established by independently:

3. Alternate experimental approaches and/or other scientific or technical innovations (“alt-GoF”

GoF RBA Report Gryphon Scientific, LLC 14

GoF RBA Report Gryphon Scientific, LLC 15

The SARS- and MERS-coronaviruses 26

An overview of GoF Research 34

GoF RBA Report Gryphon Scientific, LLC 16

4.1.1 Biology of Influenza

4.1.1.2 Virus Structure

GoF RBA Report Gryphon Scientific, LLC 17

4.1.1.3 Antigenic Variation

GoF RBA Report Gryphon Scientific, LLC 18

4.1.1.4 Host Range

GoF RBA Report Gryphon Scientific, LLC 19

GoF RBA Report Gryphon Scientific, LLC 20

16 McCandless D, Hollowood E. Influ-Venn-Za. Who can catch which flu?

GoF RBA Report Gryphon Scientific, LLC 21

4.1.2 Influenza Epidemiology22

GoF RBA Report Gryphon Scientific, LLC 22

4.1.2.1.1 Seasonal Influenza

4.1.2.1.2 Pandemic Influenza

4.1.2.2 Infectious Period